effets de dopage, de réduction de taille et d'interface

HAL Id: tel-00952979https://tel.archives-ouvertes.fr/tel-00952979

Submitted on 28 Feb 2014

HAL is a multi-disciplinary open accessarchive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come fromteaching and research institutions in France orabroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, estdestinée au dépôt et à la diffusion de documentsscientifiques de niveau recherche, publiés ou non,émanant des établissements d’enseignement et derecherche français ou étrangers, des laboratoirespublics ou privés.

Etude de matériaux pour mémoires à changement dephase : effets de dopage, de réduction de taille et

d’interfaceGiada Eléonora Ghezzi

To cite this version:Giada Eléonora Ghezzi. Etude de matériaux pour mémoires à changement de phase : effets de dopage,de réduction de taille et d’interface. Autre [cond-mat.other]. Université de Grenoble, 2013. Français.<NNT : 2013GRENY018>. <tel-00952979>

https://tel.archives-ouvertes.fr/tel-00952979

https://hal.archives-ouvertes.fr

THESE

Pour obtenir le grade de

DOCTEUR DE L’UNIVERSITE DE GRENOBLESpecialite : Physique

Arrete ministeriel : 7 aout 2006

Presentee par

Giada Eleonora Ghezzi

These dirigee par Francoise Hippertet codirigee par Sylvain Maıtrejean

preparee au sein CEA Leti, Minatec campus et LMGP (CNRS, Grenoble-INP, Minatecet de Ecole Doctorale de Physique

Material studies for advancedphase change memories: doping,size reduction and interface effect

These soutenue publiquement le 25 fevrier 2013,devant le jury compose de :

Pr. Yves BrechetProfesseur Grenoble INP, President

Pr. Olivier ThomasProfesseur Universite d’Aix-Marseille, Rapporteur

Pr. David WrightProfessor University of Exter, Rapporteur

Dr. Christophe BicharaDirecteur de recherche CNRS, Examinateur

Dr. Paola ZulianiProject leader at ST Microelectronics, Examinatrice

Francoise HippertProfesseur Grenoble INP, Directeur de these

Sylvain MaıtrejeanIngenieur-chercheur CEA, Co-Directeur de these

To my beloved family

Contents

Summary 1

Resume 5

1 Phase Change Memories Overview 9

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.2 Phase Change Memories working principle . . . . . . . . . . . . 12

1.2.1 Basic device example . . . . . . . . . . . . . . . . . . . . 15

1.2.2 Electrical conduction model . . . . . . . . . . . . . . . . 17

1.3 Physics of phase change transformations . . . . . . . . . . . . . 18

1.3.1 Amorphization . . . . . . . . . . . . . . . . . . . . . . . 18

1.3.2 Crystallization . . . . . . . . . . . . . . . . . . . . . . . 20

1.3.3 Nucleation . . . . . . . . . . . . . . . . . . . . . . . . . . 20

1.3.4 Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

1.3.5 Johnson - Mehl - Avrami - Kolmogorov (JMAK) formalism 25

1.4 Phase change materials . . . . . . . . . . . . . . . . . . . . . . . 27

1.4.1 Ge:Sb:Te compounds . . . . . . . . . . . . . . . . . . . . 27

1.4.2 Structure of crystalline and amorphous Ge2Sb2Te5 and GeTe 29

1.5 Goals and outline . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2 Effect of doping on the structure of amorphous GeTe 35

2.1 State of the art on doping effects in phase change materials . . . 36

2.2 Theory of the Pair Distribution Function (PDF) g(r) . . . . . . 40

2.3 Description of the samples . . . . . . . . . . . . . . . . . . . . . 45

2.4 X-Ray scattering measurements and results . . . . . . . . . . . . 46

2.5 Ab initio simulations . . . . . . . . . . . . . . . . . . . . . . . . 49

i

2.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

2.7 Conclusions and perspectives . . . . . . . . . . . . . . . . . . . . 61

3 Confinement of phase change materials: Ge2Sb2Te5 nanoclusters 63

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

3.1.1 Effect of shrinking size in one dimension: thin films . . . 64

3.1.2 Effect of shrinking size in two and three dimensions: nanos-

tructures . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3.2 Clusters deposition . . . . . . . . . . . . . . . . . . . . . . . . . 77

3.3 X-Ray Diffraction study . . . . . . . . . . . . . . . . . . . . . . 81

3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

3.5 Conclusions and perspectives . . . . . . . . . . . . . . . . . . . . 93

4 Interface effect on crystallization of PC thin films 95

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

4.2 Reflectivity measurements . . . . . . . . . . . . . . . . . . . . . 97

4.3 X-Ray Diffraction measurements . . . . . . . . . . . . . . . . . . 103

4.4 Synchrotron X-Ray Diffraction . . . . . . . . . . . . . . . . . . . 110

4.5 Discussion and conclusions . . . . . . . . . . . . . . . . . . . . . 112

Conclusion 121

A Experimental Techniques 125

A.1 Reflectivity measurements . . . . . . . . . . . . . . . . . . . . . 126

A.2 X-Ray Diffraction . . . . . . . . . . . . . . . . . . . . . . . . . . 127

A.2.1 Conventional X-Ray Diffraction laboratory experiment . 128

A.2.2 Large-scale facilities experiments . . . . . . . . . . . . . 129

B Deposition method 137

List of Figures 148

List of Tables 150

References 151

ii

Acknowledgments 165

iii

Summary

Phase Change Memories (PCM) are considered the best candidate for the

next generation of non volatile memories (NVM), which market is actually dom-

inated by Flash technology. PCM are based on the properties of some chalco-

genide materials, called phase change (PC) materials, to reversibly switch be-

tween a crystalline and an amorphous phase. These two phases are characterized

by very different electrical and optical properties, which makes possible to store

information. For PCM to be competitive a great research effort is still needed.

This effort should be directed to improve technological aspects such as device

architecture, layout and control circuitry, but also to optimize the PC materials

used in the memory cell. The effect of scaling and interface layers on the PC

materials properties must also be understood. Up to now the most used and

studied PC material is Ge2Sb2Te5 (GST), sometimes doped with nitrogen, but

research is extremely active in looking for other compounds that can offer better

properties. Materials to be used in PCM should crystallize fast, exhibit large

optical and electrical contrast between the amorphous and crystalline phases,

have a melting temperature sufficiently low to limit the electric power needed

for amorphization and their amorphous phase must be stable to grant good re-

tention performances. PC materials doping has been identified as a promising

solution for properties optimization. Moreover, those properties must not decay

after many transformation cycles between the amorphous and crystalline states

and must be kept with scaling, i.e. for a small amount of PC material embedded

1

in conducting and/or insulating materials. In this context, the aim of this work

is to investigate the effect of scaling and doping on PC materials properties.

Here, three directions have been explored. One is a contribution to understand

the impact of doping on the structure of amorphous GeTe. The second one is

an investigation of the effect of 3D confinement on the phase change in GST.

The third one is the study of influence of interface materials on crystallization

of GST and GeTe thin films.

In the first part of the thesis the local structural properties of C and N

doped amorphous GeTe are investigated through X-ray scattering experiments

performed at the synchrotron SOLEIL (Saclay, France). At the beginning of this

work it was demonstrated that C and N doping improves drastically the data

retention of GeTe and lowers the current needed for amorphization. The goal

was to understand the impact of doping on the amorphous structure of GeTe

by analyzing the pair distribution function of doped and undoped samples. The

impact of doping is revealed experimentally by the appearance of a new peak

in the pair distribution function of doped GeTe, indicating the formation of a

bond at a new distance, absent in the undoped amorphous material. Ab initio

simulations show the formation of new tetrahedral and triangular or pyramidal

environments centered on carbon or nitrogen as well as long carbon chains and a

few N2 molecules. The new peak observed experimentally corresponds to Ge-Ge

distances in the units centered on carbon or nitrogen. These structural changes

can be related to the enhanced crystallization temperature and activation energy

of C and N-doped GeTe.

The effect of confinement on the phase change in GST is the subject of the

second part of this work. The capability of PC materials to be scaled while main-

taining their properties is a fundamental requirement for further development of

PCM. There have been many studies on thin films of varying thickness. In some

cases the crystallization temperature has been reported to increase drastically

with reduced dimensions for films thinner than 10 nm, up to the point of losing

the phase change property for films thinner than 2 nm. These studies deal with

confinement in only one direction, the film thickness, but the ideal system for

studying the effect of scaling on PC materials in a memory cell (where both size

2

and interface effects play a role) is a set of nanoclusters. Indeed, the volume of

clusters is confined in three dimensions and the interface effects are enhanced

due to a larger surface/volume ratio. Nano-sized clusters of GST with an aver-

age size of around 5.7 nm (± 1 nm) in a matrix of Al2O3 have been made using a

sputtering gas phase condensation source and characterized by X-Ray diffraction

measurements performed at the ESRF synchrotron (Grenoble). At the moment,

those clusters are the smallest GST clusters ever deposited by sputtering, and

with the narrowest size distribution. These clusters have been made by a method

close to those used for PCM thin films deposition, thus giving information that

can be easily exported to device fabrication. The crystallization temperature

of clusters is around 180C, only slightly above the crystallization temperature

(155C ) of a 10 nm thin film of GST deposited under the same conditions and

embedded in Al2O3 . The crystalline phase is the cubic metastable phase of

GST for both clusters and thin film. The lattice parameter is larger in clusters

than in thin films. The lattice variation can be explained by supposing that

the surrounding rigid Al2O3 matrix exerts a tensile strain on the clusters and

that their volume during crystallization is forced to remain constant and equal

to their volume in the amorphous phase. Various effects could contribute to the

difference in crystallization temperature between clusters and the 10 nm thin

film, i.e. a composition effect, different surface to volume ratio, matrix influence,

stress or strain effects or an intrinsic size effect.

The third part of this thesis is dedicated to a study of the effect of interface

layers on the crystallization temperature of GeTe and GST thin film (10 to 100

nm thick). Despite its broad scientific and technological interest, this subject

has not been widely treated in literature up to now. First, measurements of the

crystallization temperature of GST and GeTe 100 nm thin films embedded in

three different materials (Ta, TiN and SiO2 ) through reflectivity measurements

are reported. It is observed that in both GeTe and GST interfaced with Ta

the crystallization temperature is around 20C higher than the one obtained by

interfacing those materials with TiN or SiO2 . Even if some studies in literature

put in evidence the influence of interface over the crystallization temperature of

Ge-Sb-Te thin films, such a remarkable interface effect in relatively thick films

3

(100 nm) was never reported before. The structural properties of the crystalline

phase of GeTe films, such as the grain size and texture, are investigated through

X-Ray diffraction analysis for the three different interface materials. The results

show that the SiO2 interfaced samples are characterized by a strong texture while

a weak texture is observed for Ta and TiN interfaced samples. Moreover, the

grain size calculated from the 012 Bragg peak width for SiO2 interfaced samples

is bigger than the ones calculated for Ta and TiN interfaces for a tilting angle

of 40of the sample. This suggest that for planes tilted by 40compared to

the (012) plane the SiO2 /GeTe surface is energetically favorable, resulting in

an abnormal growth with a preferred orientation with the (100) or (010) plane

parallel to the sample surface. If this hypothesis is true, a different nucleation

and growth mechanism for the different interfaced samples can be supposed.

In the last part, general conclusions and perspectives are presented.

4

Resume

Les memoires a changement de phase (PCM) sont l’un des candidats les plus

prometteurs pour la prochaine generation de memoires non-volatiles (NVM),

dont le marche est domine par la technologie Flash. Les PCM sont basees sur la

propriete des certains materiaux chalcogenures, appeles materiaux a changement

de phase (PC), de changer reversiblement d’etat entre une phase cristalline et une

phase amorphe. Ces deux phases sont caracterisees par des proprietes electriques

et optiques tres differentes, ce qui rend possible de stocker des informations.

Pour que les PCM soient competitives un grand effort de recherche est encore

necessaire. Cet effort devrait etre dedie d’une part a l’amelioration des aspects

technologiques, comme l’architecture des dispositifs, le layout et le circuit de

controle, et d’autre part a l’optimisation des materiaux PC utilises dans la cellule

de memoire. L’effet de la reduction de taille et celui des couches d’interface sur

les proprietes de materiaux PC doivent aussi etre compris. Jusqu’ici le materiau

PC le plus utilise et etudie est Ge2Sb2Te5 (GST), parfois dope avec de l’azote,

mais la recherche est extremement active dans l’investigation d’autres composes

qui peuvent offrir de meilleures proprietes. Les materiaux a utiliser dans les

PCM doivent cristalliser vite, montrer un grand contraste optique et electrique

entre les phases amorphes et cristallines, avoir une temperature de fusion suff-

isamment basse pour limiter l’energie electrique necessaire pour l’amorphisation

et la stabilite de leur phase amorphe doit etre grande pour garantir une bonne

retention des donnees. Le dopage des materiaux PC a ete identifie comme une

5

solution prometteuse pour l’optimisation de proprietes. De plus, ces proprietes

ne doivent pas se deteriorer apres beaucoup de cycles de transformation en-

tre les etats amorphes et cristallins et doivent etre conserves quand la taille est

reduite, c’est-a-dire pour une petite quantite de materiau PC incorpore entre des

materiaux conducteurs et/ou des isolants. Dans ce contexte, le but de ce travail

est d’examiner l’effet de la reduction de taille et du dopage sur les proprietes

des materiaux PC. Trois axes ont ete explores. Le premier est une contribu-

tion pour comprendre l’impact du dopage sur la structure de GeTe amorphe. Le

deuxieme est une investigation de l’effet de confinement en 3D sur le changement

de phase de GST. Le troisieme est l’etude d’influence du materiau d’interface

sur la cristallisation des films minces de GST et GeTe.

Dans la premiere partie de la these la structure locale de GeTe amorphe

dope avec C ou N est determinee par des experiences de diffusion des rayons X

executees au synchrotron SOLEIL (Saclay, France). Au debut de ce travail il

avait ete demontre que le dopage par C ou N de GeTe ameliore spectaculaire-

ment la retention de donnees et baisse le courant necessaire pour l’amorphisation

dans les dispositifs. Le but etait donc de comprendre l’impact du dopage

sur la structure amorphe de GeTe en analysant la fonction de distribution

de paires d’echantillons dopes et pas dopes. L’impact du dopage est revele

experimentalement par l’apparition d’un nouveau pic dans la fonction de dis-

tribution de paires de GeTe dope, indiquant la formation d’une liaison a une

nouvelle distance, absente dans le materiau amorphe non dope. Des simula-

tions ab initio montrent la formation de nouveaux environnements tetraedriques,

triangulaires ou pyramidaux centres sur le carbone ou l’azote ainsi que des

longues chaınes de carbone et quelque molecules d’N2. Le nouveau pic observe

experimentalement correspond aux distances Ge-Ge dans les unites centrees

sur le carbone ou l’azote. Ces changements structurels peuvent etre relies a

l’augmentation de la temperature de cristallisation et de l’energie d’activation

du GeTe dope C ou N.

L’effet de confinement sur le changement de phase de GST est le sujet de

la deuxieme partie de cette these. La capacite des materiaux PC a etre con-

fines en maintenant leurs proprietes est une exigence fondamentale pour le

6

developpement des PCM. Il y a eu beaucoup d’etudes sur des films minces

d’epaisseur variable. Dans quelques cas, une tres forte augmentation de la

temperature de cristallisation a ete trouvee pour des films plus minces que 10

nm, jusqu’au point de perdre la propriete de changement de phase pour des films

plus minces que 2 nm. Ces etudes traitent du confinement dans une seule direc-

tion, l’epaisseur de film, mais le systeme ideal pour etudier l’effet de la reduction

de taille sur des materiaux PC dans une cellule de memoire (ou tant la taille que

les effets d’interface jouent un role) est un ensemble de nanoparticules. En effet,

la taille des particules est limitee dans trois dimensions et les effets d’interface

sont augmentes en raison d’un plus grand rapport surface/volume. Des agregats

de GST, avec une taille moyenne d’autour de 5.7 nm (± 1 nm) et deposees

dans une matrice de Al2O3 , ont ete fabriques par pulverisation puis deposes et

caracterises par des mesures de diffraction des rayons X faites au synchrotron

ESRF (Grenoble). A l’heure actuelle, ces particules sont les plus petites partic-

ules de GST jamais deposees par pulverisation et avec la distribution de taille la

plus etroite. Ces particules ont ete fabriquees par une methode proche de celle

utilisee pour la deposition de films minces dans les dispositifs PCM, et donc

les informations obtenues peuvent etre facilement exportees vers la fabrication

des dispositifs. La temperature de cristallisation des nanoparticules est autour

de 180C, seulement legerement au-dessus de la temperature de cristallisation

(155C ) d’un film mince de GST de 10 nm depose dans les memes conditions et

encapsulee par Al2O3 . La phase cristalline est la phase metastable cubique de

GST tant pour les nanoparticules que pour le film mince. Le parametre de maille

est plus grand dans les nanoparticules que dans le film mince. On peut expliquer

la variation de parametre de maille en supposent que la matrice Al2O3 , qui est

rigide, exerce une tension sur les nanoparticules et que leur volume pendant la

cristallisation est force de rester constant et egal au volume occupe dans la phase

amorphe. Divers effets pourraient contribuer a la difference entre la temperature

de cristallisation des nanoparticules et du film mince de 10 nm, soit un effet de

composition, la difference de rapport surface/volume, l’influence de la matrice,

des effets de contraints et deformations ou un intrinseque effet de taille.

La troisieme partie de cette these est consacree a une etude de l’effet de

7

couches d’interface sur la temperature de cristallisation de films minces de GeTe

et GST (epaisseur de 10 a 100 nm). Malgre son large interet scientifique et

technologique, ce sujet n’a pas ete largement traite dans la litterature jusqu’ici.

D’abord, la temperature de cristallisation de films minces de GST et GeTe de 100

nm encapsules dans trois materiaux differents (Ta, TiN et SiO2 ) est determinee

par des mesures de reflectivite. Il est observe que, tant dans le GeTe que dans le

GST, la temperature de cristallisation obtenue dans le cas d’interface avec du Ta

est autour 20C plus haute que celle obtenue en interfaant ces materiaux avec

TiN ou SiO2 . Meme si quelques etudes de la litterature ont mis en evidence

l’influence d’interfaces sur la temperature de cristallisation de films minces de

Ge-Sb-Te, un effet d’interface si remarquable dans des films relativement epais

(100 nm) n’a jamais ete rapporte auparavant. Les proprietes structurelles de

la phase cristalline de films du GeTe, comme la taille de grains et la texture,

sont examinees par diffraction des rayons X pour les trois differents materiaux

d’interface. Les resultats montrent que l’echantillon interface avec SiO2 est car-

acterise par une texture forte tandis qu’une texture faible est observee pour les

echantillons interfaces avec Ta et TiN. De plus, la taille de grains, calculee en util-

isant la largeur du pic de Bragg 012, pour l’echantillon interface avec SiO2 est

plus grande que celle calculees pour les interfaces Ta et TiN pour un angle

d’inclinaison de l’echantillon de 40. Ceci suggere que l’interface SiO2 /GeTe

est energetiquement favorable pour les plans inclines de 40par rapport aux

plans 012, aboutissant a une croissance anormale dans cette direction. Si cette

hypothese est vraie, on peut supposer un mecanisme de nucleation et crois-

sance different pour les echantillons interfaces avec differents materiaux. Dans

la derniere partie, des conclusions generales et des perspectives sont presentees.

8

Chapter 1

Phase Change Memories

Overview

9

1.1 Introduction

1.1 Introduction

The necessity to store informations has always been one of the basic needs of

mankind throughout its history. In the last few decades, with the explosive de-

veloping of electronics and computers, this need became an urgency. Since the

formulation of Moore’s Law in 1965 [1] the microelectronics industry develop-

ment has been ruled by the trend of reducing device cost of one half every two

years. This has been achieved by conventional CMOS device architectures by

increasing the integration density of devices, meaning reducing their dimensions

or even developing new strategies as 3D device integration that allows higher

density at the same device size. Floating gate non-volatile memories (NVM),

usually named Flash memories, represent the mainstream in the NVM market.

They have been the reference technology for years, but their further scaling be-

come difficult due to technological and physical constraints. Already over the

past few years, Flash memories have faced hard challenges for keeping the scal-

ing trend and great difficulties arise for the next technology nodes that make

hard to even maintain actual specifications.

As a consequence, there is a rising industrial interest for emerging alterna-

tive NVM technologies that can offer better scaling possibilities with even better

memory performances than Flash memories. Among those alternatives the most

interesting one is the Phase-Change resistive memory (PCM). PCM are based on

the properties of some chalcogenide materials, called phase change (PC) mate-

rials, to reversibly switch between a crystalline and an amorphous phase. These

two phases are characterized by very different electrical and optical properties

and this makes possible to store the information in terms of ’0’ and ’1’ levels,

as it will be described in details in section 1.2. Phase-change memories, also

called Ovonic Unified Memory, offer a scalability beyond Flash technology and

can potentially be better than Flash memories in terms of faster random access

time, higher density and higher endurance [2].

The first to investigate extensively the switching properties of chalcogenides

and to demonstrate their practicality was S. R. Ovshinsky, during the 1950s and

1960s [3]. In early 1970s the interest in PCM rose and a first 256-bit memory

10

1.1 Introduction

array of phase change memory cells was developed by R. G. Neale, D. L. Nelson

and G. E. Moore in 1970. However, the programming operation parameters of

this memory were very poor. Other studies on phase change memory electrical

devices have been done until 1978 but the enormous power required for pro-

gramming constituted a strong limitation and they were abandoned (Chapter

1 of Ref. [4]). The reason for this enormous power consumption were the big

dimensions of the memory cell. The PCM devices were not as competitive as the

metal oxide semiconductor memories that were rising in market at those times,

so the interest in PCM subsided and there were no further developments until

the first years of 2000.

However, during the 1970s and 1980s chalcogenide materials continued to

attract interest because of their suitability for use in optical memories. The

most important result of this effort was the discovery in 1987, by Yamada and

co-workers, of the GeTe − Sb2Te3 pseudobinary line [5] and in particular of

the Ge2Sb2Te5 compound, called GST [6]. The properties of those materials

opened the way to the development of phase change optical recording devices

such as Compact Disk (CD), Rewritable Compact Disk (CD-RW), Digital Ver-

satile Disk (DVD) and Blue-Ray DVD. On the other hand, at the beginning of

2000s the new shrinking possibilities obtained through the great development

of lithographic techniques triggered a new interest in phase change memory

electrical devices [4]. The now reduced available volume of the memory cell,

resulting in lower power consumption, led PCMs to be developed by the Re-

search and Development departments of many industrial companies as Intel,

STMicroelectronics and Samsung. Nowadays, PCMs are one of the best can-

didates to replace market-leader Flash Non Volatile Memories (NVM) that are

approaching their critical size limit.

One of the open research subjects in the PCM field is material study and

optimization. While the device architecture faces the challenges of scaling and

vertical integration [2], new materials that can offer better performances should

be investigated. A deeper understanding of the phase change mechanism and

how PC materials structures are related to their properties is mandatory. Nowa-

days, this need is combined with the rising scientific interest of Phase Change

11

1.2 Phase Change Memories working principle

Figure 1.1: Basic principle of the phase transformation [8]. PC materials can switch re-

versibly between an amorphous state, corresponding to the logical level ’0’ or RESET, and a

crystalline state corresponding to the logical level ’1’ or SET. The SET operation consists in

programming the cell into the SET state, while the RESET operation consists in programming

the cell into the RESET state. To obtain the amorphous phase the PC material must be an-

nealed above its melting temperature and then rapidly cooled down. To obtain the crystalline

phase the material must be annealed above its crystallization temperature Tx .

Materials. In the last years the importance of the basic research on PC mate-

rials has been confirmed by an increasing interest of the scientific community,

and by the success of dedicated symposium as, for example, the Phase Change

symposium in the Material Research Society Spring conference. The number of

articles dedicated to studies of the fundamental properties of PC materials has

rose in the last years (see for example Ref. [7] and [8]), confirming that this is

an active and promising research field.

The aim of this thesis is to study those materials, focusing on the effect of

doping well known compounds, the effect of different material interfaces and the

shrinking effect on very small nanoparticles. The following of this first Chapter

will be dedicated to an introduction on PCM basic working principle (Section

1.2), the description of the theory of amorphization and crystallization (Section

1.3) and a brief review of the main structural characteristics of PC materials,

focusing on the structure of the well known compounds GeTe and GST (Section

1.4).


Phase change memories (PCM) are based on the property of so called phase

change materials (PC materials) to change reversibly between an amorphous

and a crystalline state, as schematically shown in Figure 1.1.

12


The phase change is obtained through heating of the material. In memory

devices the heating can be provided by electric or laser pulses. If a PC material

is heated above its melting temperature Tm and cooled down quickly (with a

cooling rate of 109 − 1011K/s) it solidifies in a glassy structure, the amorphous

phase. The glass is in a metastable state so it will tend to crystallize on very

long time scale. This time should be of the order of several years since it deter-

mines the capability to maintain the information. If the amorphous material is

annealed for a sufficiently long time (usually tens of nanoseconds) below Tm but

above its crystallization temperature Tx , it switches to the crystalline phase.

The theory of the phase change mechanism will be explained with more details

later in Section 1.3. The two phases are characterized by very different optical

and electrical parameters, thus providing the contrast required to distinguish

between logical states. For example, the amorphous phase exhibits a high value

of resistivity and a low value of reflectivity, and vice-versa for the crystalline

phase. The optical contrast between the amorphous and crystalline phase is

illustrated in Figure 1.2 for the case of Ge2Sb2Te5 .

The phase change property allows to store an information by associating the

logical level ’0’ and ’1’ to the two different phases. Traditionally, the level ’0’

(or RESET level) has been associated to the amorphous phase and the level

’1’ (or SET level) to the crystalline phase. Crystallization is the slowest process

involved. It must occur quickly in the programming operation in order to achieve

fast programming speed, but the spontaneous crystallization of the metastable

amorphous phase should not take place for many years at room temperature in

order to grant data retention. This means that the crystallization rate of PC

materials must increase by orders of magnitude with the change in temperature

between room temperature and Tx .

In the following sections the working principle of a schematic cell device will

be explained and the programming curves and electrical characteristics of the

cell will be presented.

13


Figure 1.2: Evidence of the different optical properties of the PC material Ge2Sb2Te5 in the

amorphous and crystalline phases. The reflectivity of GST is reported as a function of tem-

perature starting from an initially amorphous sample. The amorphous phase is characterized

by a low reflectivity value compared to the one of the crystalline phase. On the graph it is

easy to identify the crystallization temperature at which the phase transformation occurs.

14


Figure 1.3: Schematic representation of the lance-like structure of a PCM cell device. The

PC material is interfaced with a top electrode and a bottom electrode (heater).

1.2.1 Basic device example

The schematic picture of a Ovonic Unified Memory (OUM) PCM cell in its

simplest form (lance-like structure) is reported in Figure 1.3.

The device consists in a thin film of PC material which is electrically acces-

sible by a top electrode and a bottom electrode, also called heater . For almost

all the PC materials integrated in devices the as-fabricated cell is entirely crys-

talline, due to the high temperatures reached in fabrication process. In order

to read and program the cell, an imposed external voltage is applied at the

electrodes generating a current that flows from to the top electrode the heater

through the PC material. To read the state of the cell a low power current pulse

is imposed and the overall resistance of the cell is measured. If the resistance is

high the cell is in the RESET state, while if the resistance is low the cell is in

the SET state. It is worth noting that the cell is considered to be in the SET

state when the read resistance value is sufficiently low, and this can be achieved

by the formation of crystalline paths percolating through the amorphous volume

and not necessarily by crystallizing the entire volume. Concerning the program

operations, the shape and intensity of the applied current pulse determines if

the cell will be programmed in the amorphous state (RESET operation) or in

15


Figure 1.4: Current pulses for the programming operation of the cell. RESET pulse (a)

SETMIN pulse (b) and SET pulse (c).

the crystalline state (SET or SETMIN operation) as shown in Figure 1.4. The

contact area between the PC material and the heater is very narrow so the

current density reaches its highest value at the PC material - heater interface.

As shown in Figure 1.4a, for the RESET operation a current pulse of high in-

tensity with a rapid falling edge is applied (RESET pulse) so that the current

density at the PC material / heater interface is sufficiently high to heat the

PC material over the melting temperature by Joule heating. The abrupt falling

edge of the RESET pulse induces a fast quench of the material that solidifies in

the amorphous phase in an hemispherical volume. Amorphization of this area

blocks the low-resistive current path and results in an overall large resistance.

It is worth underlining that reducing the heater dimension, meaning reducing

the cell size, results in an increased current density so that the current required

for amorphization is reduced. The SET pulse can be chosen to have the same

shape as the RESET pulse but with a lower intensity and a longer duration

(SET pulse, Figure 1.4b) or an intensity sufficiently high to melt the PC mate-

rial but with a slow falling edge (SETMIN pulse, Figure 1.4c). In the first case

the PC material is heated below its melting temperature and the amorphous to

crystalline transition is induced. Usually the transformation does not involve

all the amorphous volume but results in the creation of percolating paths, as

explained before. Thus the overall resistance is higher than the resistance of an

entirely crystalline cell, but still two or three order of magnitude lower than the

RESET value. In the SETMIN case the PC material is first melted and then

crystallized by a slow quenching and the final resistance value is lower than in

the SET case.

16


Figure 1.5: I-V characteristic of a PCM cell in the crystalline and amorphous states (from

Ref.[9]). The I-V characteristic of the amorphous state present a snap-back in correspondence

of a threshold voltage that is not present in the crystalline I-V curve.

1.2.2 Electrical conduction model

The mechanism of electrical conduction in phase change memories is a very ac-

tive research field, due to the important role of electrical properties in devices.

The current-tension (I-V) characteristic of a PCM cell is reported in Figure 1.5.

It is possible to distinguish between a low electrical field region and an high elec-

trical field region. For low applied voltages (low electrical field region) the GST

amorphous conductance is low compared to the crystalline GST conductance.

When the external bias reaches a certain value (called threshold switching volt-

age) a snap-back takes place and the conductance abruptly switches to a higher

value. This threshold switching, also called Ovonic Threshold Switching (OTS),

was first discovered by Ovshinsky [3]. The crystalline GST I-V curve presents

no evidence of such a switching and in the high field region the conductances

of both states are equal. It is important to underline that the threshold switch-

ing does not correspond to the amorphous to crystalline transition. After the

switching takes place, the cell remains amorphous until Joule heating is suffi-

cient for inducing crystallization. The OTS is fundamental in order to grant

low power dissipation during the SET operation. When the voltage applied to

the amorphous cell exceeds the threshold value, the current flowing through it

17

1.3 Physics of phase change transformations

increases drastically, allowing the phase transition to take place with a low ap-

plied power. Without the switching phenomenon, the high power required to

perform the SET operation would make the programming operation unpracti-

cal. The physical mechanism for OTS has been widely investigated. Even if it is

still not conclusively clarified, it is generally supposed to be an electronic effect

rather than a thermal or structural effect. Ielmini and Zhang proposed a model

for conduction [10, 11] described by the Poole-Frenkel effect that well describes

the transport properties of the crystalline and amorphous phases as well as the

switching effect.


1.3.1 Amorphization

As already stated in the introduction, the amorphous phase can be obtained from

the melt by a rapid quench, fast enough to avoid crystallization. If a liquid is

cooled below its melting temperature Tm it does not crystallize instantaneously

and it can be undercooled. While the temperature decreases, the liquid viscosity

η increases. Such an undercooled liquid is in a metastable equilibrium, meaning

that it is metastable with respect to the crystalline stable phase but it is still

in its internal equilibrium. If the undercooled liquid is cooled down below the

so-called glass transition temperature Tg it becomes configurationally frozen,

losing the thermal equilibrium, and it becomes a glass. Tg is commonly defined

as the temperature at which the time scale necessary for atomic rearrangements

becomes larger than the measurement time, and it usually occurs at the point

where the viscosity equals 1 × 1012 Pa s [12, 8]. Below Tg the glass has a very

low microscopic atomic mobility D(T ), which is inversely proportional to the

macroscopic viscosity η(T ) according to the Stokes-Einstein relation D(T ) ∝

T/η(T ).

Crystallization is thermodynamically forbidden above Tm and it is extremely

slow below Tg , while it can rapidly occur for temperatures between Tg and

Tm. For temperatures slightly below Tm the driving force for crystallization

is so low that crystallization would occur only very slowly, so with a rapid

18


Figure 1.6: Time-temperature-transformation (TTT) diagram for a PC material taken from

Reference [8]. The phase transformation of a fixed volume of PC material is reported depending

on the time spent at a certain temperature. The two orange lines on the graph indicate two

different constant rate quenching processes while the two purple lines indicate two annealing

processes starting at room temperature.

quenching of the liquid below Tg it is possible to avoid crystallization and result

in the formation of the amorphous phase. This is illustrated in Figure 1.6a,

reported from Ref. [8], in which the phase transformation is shown in terms of a

time-temperature-transformation (TTT) diagram. The corresponding mobility

and the driving force for crystallization are reported in Fig. 1.6b and 1.6c,

respectively. In the TTT diagram the phase transformation of a fixed volume of

PC material is reported, depending on the time spent at a certain temperature.

The two orange lines on the graph indicate two different constant rate quenching

processes. If the cooling is sufficiently fast (around 109K/s) crystallization can

be avoided slightly below Tm due to the very low driving force for crystallization

(Figure 1.6c). With further fast undercooling, the driving force increases but

the mobility decreases and if Tg is reached the material becomes amorphous.

On the other hand, if the cooling rate is too slow the material will crystallize.

The amorphous phase can crystallize below Tg only for very long times. This

time has a very important role in determining the data retention in a memory

cell, as it defines the time at which a cell in the RESET state will crystallize

losing the stored information. On the TTT graph are also reported in purple two

annealing processes starting at room temperature, showing that crystallization

time and temperature are dependent.

19


1.3.2 Crystallization

Two different process contributes to crystallization of an amorphous solid.The

first one is the nucleation, that initiates the crystallization through the formation

of small crystalline nuclei. The second one is the growth of those nuclei to

a macroscopic size. The so-called classical nucleation theory of steady state

nucleation has been developed by Volmer, Weber, Becker, Doring, Turnbull and

Fisher during the first decades of the 20th century [13, 14, 15, 16], based on the

pioneering work of Gibbs [17].

1.3.3 Nucleation

Nucleation can occur in two different ways. In the first and simplest case, called

homogeneous nucleation, the crystallite germinates inside the amorphous phase,

without involving other substances. If instead the amorphous phase is in contact

with other substances that act as preferred sites for nucleation, an heterogeneous

nucleation occurs.

Homogeneous Nucleation

Form Gibbs’ thermodynamical theory, clusters of radius r can be formed inside

the amorphous phase by thermodynamics fluctuations. Their equilibrium distri-

bution is ruled by the Boltzmann statistic, meaning that the number of clusters

of radius r at equilibrium is

N equ(r) = N0 · exp(−∆Gcluster(r)

kBT) (1.1)

where N0 is the total number of atoms in the liquid, ∆Gcluster(r) is the reversible

work for crystal cluster formation, kB is the Boltzmann constant and T is the

absolute temperature. ∆Gcluster(r) can be written as

∆Gcluster(r) = −∆Glc,V ·4

3πr3 + 4σπr2 (1.2)

where ∆Glc,V is the Gibbs free energy difference per volume between the crys-

talline and the amorphous phase and σ > 0 is the interfacial free energy. ∆Glc,V

is zero at Tm and positive for T <Tm . So Eq. 1.2 is composed of a nega-

tive volume term that becomes larger as the temperature T is reduced below

20


Figure 1.7: Evolution of ∆Gcluster(r) as a function of r corresponding to Eq. 1.2, taken

from Chapter 7 of Reference [4]. The curve exhibit a maximum for the r = rc (critical radius)

that corresponds to the critical work for cluster formation ∆Gc.

Tm and a surface term, always positive, that results from the creation of a clus-

ter/liquid interface. The evolution of ∆Gcluster(r) as a function of r is depicted

qualitatively in Figure 1.7.

The curve present a maximum at

rc =2σ

∆Glc,V

(1.3)

where the critical radius rc is of the order of a few nanometers. The energy

of a nucleus of radius rc may be calculated by substituting rc to r in 1.3, thus

obtaining the critical work for cluster formation ∆Gc

∆Gc = ∆Gcluster(rc) =16π

3

σ3

(∆Glc,V )2. (1.4)

Clusters with radius r = rc are the so-called critical clusters. The evolution

to bigger dimensions of clusters with r < rc is energetically not favorable so

they spontaneously decay, while clusters with r > rc can grow due to the free

energy gain. This means that ∆Gc constitutes a barrier against crystallization,

the same barrier that impedes immediate crystallization of the amorphous phase

when it is undercooled below Tm .

The approach of Gibbs is purely thermodynamic, and on its basis a first

kinetic model for nucleation was developed by Volmer and Weber [13, 14]. They

21


modified Eq. 1.1 by taking into account the fact that it becomes unphysical

for r > rc, where the clusters distribution at equilibrium begin to increase with

increasing r. To avoid this, N equ(r) was set to zero for r > rc. By considering

that the nucleation occurs when a critical cluster gains one more atom, the

nucleation rate was calculated per unit volume per second

Iequ = sc · k ·N equ(rc) = sc · k ·N0 · exp(−∆Gc

kBT) (1.5)

where k is the arrival rate to the crystalline cluster of amorphous phase atoms

and sc is the number of surface atoms of the cluster.

One of the limitation of the Volmer-Weber theory is that a critical cluster

that gain one more atom is supposed to grow to macroscopic size while in reality

there is still a probability for it to decay. Backer and Doring [13, 14, 15] proposed

a different expression for the equilibrium cluster distribution N equ(r) that takes

into account that possibility, thus obtaining the following steady state nucleation

rate

Iss = sc · k ·N0 ·1

ic· (

∆Gc

3πkBT)

︸︷︷︸

ΓZ

·exp(−∆Gc

kBT) = Iequ · ΓZ (1.6)

where ic is the number of atoms in the critical clusters and ΓZ is the so-called

Zeldovich factor, which has been found to be usually between 1/100 and 1/10.

The weak dependence of ΓZ on temperature, especially if compared with the

exponential term, makes Eq. 1.6 essentially equal to Eq. 1.5 for most practical

purposes, but in the case of Eq. 1.6 the kinetic problem has been correctly

treated.

Up to the Volmer-Weber model, all the results were obtained by considering

a gas as a amorphous phase. In this case the value of the arrival rate k was calcu-

lated from the gases theory. The value of the pre-exponential factor of 1.6 for an

amorphous material was calculated by Turnbull and Fisher [16], who completed

the classical nucleation theory. They distinguished between a diffusion-limited

crystallization and a collision-limited crystallization. The former is the case of

phase change materials, and the expression for k is

k =6D

λ2(1.7)

22


where λ is the average interatomic distance. By using the Stokes-Einstein

equation that relates the diffusion coefficient D and the viscosity η it is possible

to write

ηD =kBT

3πλ, (1.8)

and the nucleation rate Iss for the diffusion-limited model can be expressed as

a function of η:

Iss = sc ·2kBT

ηπλ3·N0 · ΓZ · exp(−

∆Gc

kBT) (1.9)

The pre-exponential factors of Iss can be estimated in both cases by sub-

stituting reasonable values for N0, sc T and ΓZ . They result to be 1036

ηfor the

diffusion-limited case and 1039 for the collision-limited case, with an uncertainty

of two to four orders of magnitude. This has not a great influence on the overall

expression due to the strong dependence of Iss on ∆Gc and σ, both present

in the exponential term. The diffusion-limited Iss tends to zero near Tm and

Tg and exhibit a maximum for a temperature intermediate between them, as it

happens for phase change materials. This is not the case for collision-limited

kinetics, where Iss increases continuously as the temperature decreases below

Tm .

Heterogeneous Nucleation

The model for heterogeneous nucleation was developed by Volmer and Weber

[18]. It is based on the Gibbs’ model already described for the homogeneous nu-

cleation, but considering a flat substrate that act as a heterogeneous nucleation

site.

In Figure 1.8 is reported the model for heterogeneous nucleation, taken from

Chapter 7 of Reference [4]. In this model the crystalline cluster nucleates on the

heterogeneous substrate as a spherical cap of radius r. This spherical cap can

be considered as the exposed part of a complete sphere of radius r, so that the

fraction of the exposed volume can be calculated as a function of the wetting

angle θ

f(θ) =(2 + cosθ)(1− cosθ)2

4. (1.10)

23


Figure 1.8: Model for the heterogeneous nucleation taken from Chapter 7 of Reference [4].

The crystalline cluster is a spherical cap which correspond to the exposed part of a sphere

of radius r. In the schematic picture are also reported the wetting angle θ and the crystal-

substrate, amorphous-substrate and amorphous-crystal interfacial energies (respectively σcs,

σls and σlc).

It was demonstrated by Volmer and Weber that the free energy for cluster

formation ∆Gcluster is reduced if

σcs − σls < σlc (1.11)

where σcs, σls and σlc are the crystal-substrate, parental phase-substrate and

parental phase-crystal interfacial energies, respectively. In this case, ∆Gcluster

for the heterogeneous nucleation is the ∆Gcluster for the homogeneous nucleation

multiplied by the factor f(θ)

∆Ghetc = ∆Ghet

c · f(θ). (1.12)

The critical radius rc remains unchanged, and the whole model for homo-

geneous nucleation is still valid with the only difference of a lower ∆Gcluster.

However, the parent phase atoms that can act as nucleation sites are not all the

atoms in the parent phase but only those interfaced with the substrate, so their

number is decreased. If ǫ is the fraction of parent phase atoms that are in con-

tact with the substrate on the total, the ratio of homogeneous and heterogeneous

24


nucleation rates is

Iss,het

Iss,hom= ǫ · exp(

∆Gc

kBT· [1− f(θ)]). (1.13)

The heterogeneous nucleation rates have been observed to be much higher than

the homogeneous ones, implying that θ must be small.

1.3.4 Growth

When a cluster has reached its critical radius it grows to a macroscopic size.

This growth is interface-controlled by the addition of new parental phase atoms

in the crystalline cluster [13]. The crystal growth velocity is

u = γs · λ · k · [1− exp(−∆Glc,atom(T )

kBT)](T < Tm) (1.14)

where 0 < γs < 1 is the fraction of sites on the interface where a new parent phase

atom can be incorporated, λ is the average interatomic distance, ∆Glc,atom(T )

is the Gibbs free energy between the parent phase and the crystalline phase per

atom and k has the same meaning and value as for nucleation. For diffusion-

limited kinetics, by substituting k as in Section 1.3.3 and using Eq. 1.8 the

crystal growth velocity is

u = γs ·2kBT

ηπλ2· [1− exp(−

∆Glc,atom(T )

kBT)](T < Tm) (1.15)

As for the nucleation rate, the growth velocity u is zero at T =Tm , negligible

at T = Tg and exhibit a maximum in the temperature range between Tg and

Tm .The growth rate maximum is usually located at a higher temperature than

the nucleation rate maximum.

1.3.5 Johnson - Mehl - Avrami - Kolmogorov (JMAK)

formalism

The so-called JMAK (Johnson-Mehl-Avrami-Kolmogorov) model is an alterna-

tive to the classical crystallization theory for describing the crystallization ki-

netics. While the classical nucleation theory allows the calculation of the cluster

nucleation and growth rates, as well as their size distribution, the JMAK model

25


is a mean to calculate the crystalline fraction in terms of crystal nucleation and

growth rates.

The model is based on the JMAK equation, which gives the volume frac-

tion of the transformed material as a function of time (x(t)) under isothermal

annealing conditions:

x(t) = 1− exp (−ktn) (1.16)

where t is time, k is an effective rate constant and n is the so called Avrami coef-

ficient. The value of k depends on temperature through the Arrhenius equation

k(T ) = νexp

(EA

−kBT

)

(1.17)

where ν is the frequency factor, EA is the activation energy, T is the absolute

temperature and kB is the Boltzmann constant. The value of ln [−ln (1− x)]

plotted as a function ln(t) is the so-called JMAK plot. In literature, the JMAK

theory has been often used to interpret isothermal annealing of PC materials,

and the activation energy EA has been usually determined through Kissinger

analysis [19]. The Kissinger method is based on the measurement of the variation

of the crystallization temperature Tx for different heating rates dT/dt, which are

related to the activation energy through the equation

ln

(1

T 2x

·dT

dt

)

= −EA

kBTx

+ C (1.18)

so that it is possible to deduce the activation energy as the slope of the linear in-

terpolation of the plot of ln(

1T 2x· dT

dt

)

versus 1kBTx

. However, even if this method

is widely used, it is based on the assumption of an Arrhenius-like temperature

dependence for crystallization. This is not the case when the crystallization is

controlled by the nucleation rate, which is non-Arrhenius [20].

Eq. 1.16 can be applied under the conditions that nucleation occurs ran-

domly and uniformly with a time independent rate and that growth is interfaced-

controlled and with a size independent rate. The use of Eq. 1.16 should not

be legitimate without fulfilling those conditions, which are usually not verified

for GST, and as a consequence the values reported in literature for EA and

ν differ significantly one from each other. However, due to its simplicity the

Kissinger method will be used in Chapter 3 and 4 of this thesis for quantitative

comparisons between materials.

26

1.4 Phase change materials

Required property of PC material Specification

High-speed phase transition Induced by nanosecond laser or voltage pulse

Long thermal stability of amorphous state At least several decades at room temperature

Large optical change between the two states Considerable difference in refractive index or ab-

sorption coefficient

Large resistance change between the states Natural consequence of the phase transformation

Large cycle number of reversible transitions More than 100,000 cycles with stable composition

High chemical stability High water-resistivity

Table 1.1: Properties that characterize PC materials [7].


Materials to be used in PCM should meet several strict requirements. They

should possess the properties of a fast crystallization, a large optical and elec-

trical contrast between the amorphous and crystalline phases, a melting tem-

perature sufficiently low to limit the electric power needed for amorphization

and a high stability of the amorphous phase to grant good retention perfor-

mances. Moreover, those properties must not decay with cycling between the

states. The crucial properties of phase-change alloys are summarized in Table

1.1, taken from Reference [7].

1.4.1 Ge:Sb:Te compounds

As already described in Section 1.1, from the material point of view the greatest

discovery for phase change memories was done in the 1980s by Yamada and

his coworkers. They identified the materials belonging to the GeTe − Sb2Te3

pseudo-binary line in the Ge:Sb:Te ternary phase diagram as the ones with the

best properties. In Figure 1.9 the Ge:Sb:Te phase diagram is represented, with

several PC materials and the GeTe− Sb2Te3 pseudo-binary line. In particular,

the Ge2Sb2Te5 compound (simply called GST) became a standard for optical

storage devices that have been developed during the 1990s. It was chosen for

its high retention time, fast transformation speed and large optical contrast be-

tween crystalline and amorphous phase. Thus, GST was chosen as the active

27


Figure 1.9: PC materials reported on the ternary Ge:Sb:Te phase diagram, with the GeTe−

Sb2Te3 pseudo-binary line put in evidence (taken from Ref. [8]).

material to be first employed in PCM due to the wide number of studies already

performed on it. However, the study of alternative phase change materials that

can offer better properties for PCM is very active and rich, both for its scien-

tific interest and the industrial development. For example, in 2011 Cheng and

coworkers studied compounds along the GeTe−Sb line. They demonstrated that

Ge-rich Ge2Sb1Te2 (the so-called golden composition) can offer better proper-

ties than GST in terms of crystallization speed and data retention [21]. It has

been also recently shown that GeTe compound can be a good candidate for em-

bedded PCM due to its higher crystallization temperature and better retention

time compared to GST [22]. The main properties of Ge2Sb2Te5 and GeTe are

compared in Table1.2

Phase change materials that belong to the GeTe − Sb2Te3 pseudo-binary

line are characterized by a few typical structural motifs indicating a common

bonding mechanism that could account for their properties. In general, ternary

compounds exhibit a rocksalt-like structure in their crystalline phase where the

anion sublattice is occupied by atoms of Te and the cation sublattice is randomly

occupied by Ge, Sb or vacancies [7].

The constant feature for all crystalline Ge:Sb:Te phase change alloys is the

presence of a more or less distorted octahedral-like coordination resulting from

a Peierls distortion. In addition, some materials are characterized by a consid-

28


Properties Ge2Sb2Te5 GeTe

Crystallization temperature Tx [22] 145 C 185C

Activation energy EA (thin films) [23] 2.3 eV 2.0 eV

Activation energy EA (devices) [24] 3.13 eV 3.2 eV

Crystalline phase [25] Rocksalt cubic (with vacancies) Rhombohedral

Lattice parameters [25] a = 6.01 A a = 4.16 A, c = 10.69 A

Density (amorphous phase) [25] 5.86 g/cm3 5.60 g/cm3

Density (crystalline phase) [25] 6.13 g/cm3 6.06 g/cm3

Table 1.2: Comparison between the main properties of Ge2Sb2Te5 (GST) and GeTe.

erable number of vacancies (for example, 20% for Ge2Sb2Te5 [25] and 25% for

Ge1Sb2Te4 [7]). The amorphous phase differs considerably from the crystalline

one by the absence of a long range order, but a local order still exists in the amor-

phous structure of PC materials. It has been investigated both experimentally

and through ab-initio calculations.

In the following section the amorphous and crystalline structure of GeTe and

GST compounds will be described.

1.4.2 Structure of crystalline and amorphous Ge2Sb2Te5 and

GeTe

Both GeTe and GST are characterized by two different crystalline phases.When

amorphous Ge2Sb2Te5 (GST) is heated, it crystallizes at around 150C in a

metastable crystalline phase with a fcc rocksalt structure. The crystalline struc-

ture of Ge2Sb2Te5 (GST) has been described in 2000 by Yamada and coworkers

[26] as it is shown in Figure 1.10. The structure is a NaCl rocksalt structure with

an octahedral-like atomic arrangement, where the Te atoms occupy one lattice

site and the Ge and Sb atoms randomly occupy the second lattice site. The Te

sites are all fully occupied, while the Ge/Sb sublattice is characterized by the

presence of around a 20% of vacancies so that the structure is characterized by

local distortions. If cubic GST is heated above around 200C it transforms in

an hexagonal phase [6].

29


Figure 1.10: Structure of GST in its crystalline metastable phase. One sublattice is occupied

by Te atoms (light blue) while the other is randomly occupied by Ge or Sb atoms (dark blue)

or vacancies (around 20% ). The cubic lattice parameter is 6.03 A [26].

30


Figure 1.11: Structure of crystalline GeTe in its rhombohedral phase. The structure can be

described as a rocksalt-like structure, distorted by a relative shift of the sublattices along the

[111] direction. It is characterized by long (3.127 A) and short (2.87 A) Ge-Te bonds shown

respectively in white and green.

Amorphous GeTe crystallizes at 180C into a rhombohedral phase (space

group R3m) that is the stable crystalline phase at room temperature [25]. Above

around 430C this phase transforms into a cubic fcc rocksalt structure (space

group Fm3m) where Ge occupies one sublattice and Te the other sublattice. The

low temperature rhombohedral phase can be described as a slightly distorted

rocksalt structure obtained by a relative shift of both sublattices along the [111]

direction, so that each atom form three short bonds and three long bonds with

its nearest neighbors. The structure is represented in In Fig.1.10, and it can be

easily observed that each atom is in a distorted octahedral environment.

The structure of the amorphous GeTe and GST phases have been widely

studied in literature [27, 28] but in the following the subject will be treated

only briefly. The short and medium range order has been studied in litera-

ture mainly through X-ray scattering and measurement of the pair distribution

function method (see Chapter 2) that allow to determine the local atomic en-

vironments. It has been observed that a local chemical ordering takes place in

the amorphous phase by an alternation of Te and Ge(or Sb) atoms which is a

31

1.5 Goals and outline

precursor of the order in the crystalline phase. The first Te-Ge distance has been

confirmed to be around 2.6A, but the Ge coordination number is still debated.

Some studies through extended X-ray absorption fine structure (EXAFS) and

X-ray absorption near-edge spectroscopy (XANES) conclude that germanium is

tetrahedrally coordinated [29]. They were further supported by X-ray fluores-

cence and Raman scattering experiments [30, 31, 32]. However, further studies

with X-ray diffraction data and reverse Monte Carlo simulations indicate for

GST bond angles around 90and no homopolar bonds, while only for GeTe the

presence of both homopolar Ge-Ge bond and a deviation in bond angles from

90was observed [27, 28, 33]. In conclusion, the chemical alternation of Te and

Ge(Sb) atoms is confirmed and germanium is found to coexist in both tetrahe-

dral and distorted octahedral environments.

For both GeTe and GST, the density of the crystalline phase is higher than

the one of the amorphous phase, so the volume is reduced after crystallization,

and the density change for GeTe is relatively higher than for GST. Amorphous

GeTe has a density of 5.60 g/cm3, while for the crystalline GeTe it is 6.06 g/cm3.

The density of amorphous GST is 5.86 g/cm3 and the one of crystalline GST is

6.13 g/cm3[25].


For PCM to be competitive it is fundamental to develop memory devices with

good retention properties, high cyclability, high transformation speed, low power

consumption and most of all high integration density, meaning cells of reduced

dimensions. Those requirements can be fulfilled both through device and lay-

out architecture improvements and through an optimization of the used PC

materials. As an example, in the field of embedded memories for automotive

systems the operating temperatures are close to the crystallization temperature

of Ge2Sb2Te5 (GST), so this material is unable to fulfill the requirements on

data retention and alternative materials must be used. Good PC materials can-

didates should have the properties listed in Table 1.1 and they must be able to

maintain these properties even when dimensions are reduced and the material

32


gets more and more confined in the cell structure.

In the last years, possible alternatives to GST have been found in using

different compounds, as GeTe, or in doping GST with N. The fast transition

speed and good retention properties of GeTe makes it suitable for embedded

applications. Right before the beginning of this thesis work in has been shown

that C and N doping improves drastically the data retention of GeTe and lowers

the current needed for amorphization. However, up to now the effects of doping

on the structure of GeTe are still not clear, in particular on the stabilizing effect

on the amorphous phase. Moreover, the effect of confinement on PC materials

is still an open subject even for standard GST. It is also important to underline

that in a memory device the active layer is always in contact with different

interface materials that can eventually influence its properties, especially for

reduced dimensions.

For those reasons, the aim of this thesis is to study PC materials proper-

ties in three directions. First, the effect of C and N doping on the structure

of amorphous GeTe has been investigated and some interesting results on the

stability of the amorphous phase have been obtained. Second, the effect of three-

dimensional confinement has been studied on very small nanoclusters of GST,

deposited by sputtering with a new technique and characterized through X-Ray

diffraction. Third, the effect on the crystallization temperature of interfacing

GST and GeTe thin films with different materials has been studied through

reflectivity and X-ray diffraction measurements.

In Chapter 2 the study on the effect of doping on GeTe will be discussed.

From X-ray scattering experiments performed at the synchrotron SOLEIL (Saclay)

on amorphous powders of GeTe, C-doped GeTe and N-doped GeTe pair distri-

bution function (PDF) will be obtained. The PDF is proportional to the prob-

ability of finding two atoms at a certain distance. This quantity provides useful

information on the local structure of a material, which is the only exploitable

one for the amorphous phase. The experimental results will be interpreted and

discussed thanks to ab-initio simulations.

The effect of confinement on GST is the subject of Chapter 3. The de-

position of nano-sized clusters of GST with an average size of around 5.7 nm

33


(± 1 nm) in a matrix of Al2O3 will be described. At the moment, those clus-

ters are the smallest GST clusters ever deposited by sputtering, and with the

narrowest size distribution. X-Ray diffraction measurement performed at the

ESRF synchrotron (Grenoble) in order to observe the crystallization of clusters

will be be reported and the discussions and conclusions will be focused on the

interpretation of the observed crystallization temperature.

In Chapter 4, the effect of interface layers on the crystallization of phase

change material will be investigated. This subject has not been largely treated

in literature up to know, so it is a new and very interesting field of research.

First, the measurements of the crystallization temperature of GST and GeTe

thin films embedded in three different materials (Ta, TiN and SiO2 ) through

reflectivity measurements will be reported. In the following, the study will be

focused on GeTe only and structure properties of the crystalline phase such as

the grain size and texture will be investigated through X-Ray diffraction analysis.

Some hypothesis will be presented on the nature of crystallization in presence

of different interfaces to support the conclusions on the obtained results.

34

Chapter 2

Effect of doping on the structure

of amorphous GeTe

35

2.1 State of the art on doping effects in phase change materials

2.1 State of the art on doping effects in phase

change materials

As already stated in the first Chapter, PCM must fulfill many requirements in

order to be competitive on the market. Those requirements include having a

good data retention and a low reset current, so that the data are preserved at

least 10 years at room temperature and the power required for programming

the cell in the RESET state is low. The data retention depends on the time

required for the metastable amorphous phase to crystallize, so it can be im-

proved by employing PC materials with a more stable amorphous phase and

a higher crystallization temperature Tx . The RESET current depends on the

resistivity of the crystalline phase because the reamorphization process requires

to heat the material by Joule effect through the programming current pulse, and

a more resistive material can be heated by lower current pulses. These goals can

be achieved by exploring new Ge-Sb-Te compound, but another possibility to

increase both the resistivity of the crystalline phase and the retention in the

amorphous state is opened by doping 1. The effect of introducing a dopant ele-

ment in PC materials has been widely investigated during the last two decades.

The most studied dopant is Nitrogen (N), but many other doping elements have

been studied lately, including Boron (B), Silicon Dioxide (SiO2 ) and, more re-

cently, Carbon (C).

The effect of N doping has been studied since late 90s, when it was observed

that cyclability for optical disks was improved by adding nitrogen into a Ge-

Sb-Te recording layer [34, 35]. When phase change electrical memories started

to be intensively studied, after year 2000, nitrogen-doped GST (NGST) gained

interest as a PCM device material due to its high crystallization temperature,

high retention time and high resistivity [36, 37]. NGST has been successfully

integrated into PCM devices [38, 39] and even in memory arrays [40]. The

literature about N-doped GST is quite vast but it will not be treated further

here, apart from a results on the structure of NGST reported in section 2.6.

1The term doping is conventionally used in this context to indicate the addition to the PC

material of elements in a concentration of several percents.

36


Figure 2.1: Low field cell resistance as a function of progam current for GST and GeTe cells

for various programming pulse times [24]. It can be noted that the SET operation for the

GeTe cell is faster and the difference in the resistance of the amorphous and crystalline phases

is higher.

As already stated in section 1.4, during the last years the GeTe compound

has been identified as a possible good candidate for embedded PCM due to its

higher crystallization temperature and better retention time than undoped GST

[22]. Moreover, GeTe PCM cells are characterized by a rapid SET operation

compared to GST cells (see Fig.2.1) and have demonstrated high cyclability

[24]. However, the required RESET current is the same for GST and GeTe.

In order to increase the resistivity of the crystalline cell and improve further

the properties of good retention and high Tx , N-doped GeTe has been studied.

N-doped GeTe (hereafter called GeTeN) with various percentages of N has been

successfully integrated into memory devices, with beneficial effects compared to

standard GeTe such as an even faster crystallization and a higher retention time.

The best performances were obtained for a N concentration of 2% [41], as shown

in Fig.2.2. An increase of the resistance of the crystalline phase for GeTeN

(N=8.4%) compared to undoped GeTe and an increase of the crystallization

temperature for GeTeN (N=9.81%) have also been observed [42, 43].

37


Figure 2.2: Calculation of the activation energy EA by interpolation of the fail times as a

function of 1/kT . In order to obtain the fail time, a PCM cell is written in the RESET state

and the fail time is defined as the time at which the resistance of the cell is reduced by one

half.

More recently, it has been shown that also C-doping can have a great impact

on the performances of GeTe. GeTeC devices with doping percentages of 4 and

10% showed an improved data retention in temperature compared to undoped

GeTe, indicating that the amorphous phase stability is improved by C-doping

[44]. As can be seen in Fig.2.3 and Table 2.1 [45], the crystallization temperature

of GeTeC is much higher than that of undoped GeTe and the effect of C-doping

is even stronger than the one of N-doping. For example, for GeTeC (C=10%)

Tx ≈ 325C , to be compared with Tx ≈ 275C of GeTeN (N=10%). Moreover,

the retention time, the activation energy EA and the crystal resistivity increase

with doping while the reset current decreases, as shown in Fig.2.4.

Doped phase change materials are thus promising candidates in order to fulfill

some requirements on PCM as the high data retention at elevated temperatures

or the low reset current. However, even if their good performances have been

demonstrated, a deeper understanding of the effects of doping on the structure

of GST and GeTe was lacking. Clarifying this subject can lead to a better

material engineering. The open questions, at least when this study has been

performed, include the impact of dopants on the amorphous structure and their

38


Figure 2.3: Reflectivity measurements of C and N doped GeTe films (150 nm thick) [45]. In

both cases Tx increases with increasing doping concentration and the effect is stronger for C

doping.

Figure 2.4: Activation energy (left) calculated for undoped and C-doped GeTe and low

electric field resistance as a function of the programming current (right) for a GST, undoped

GeTe and C-doped GeTe cell [44]. The activation energy increases and the RESET current

decreases with doping.

39

2.2 Theory of the Pair Distribution Function (PDF) g(r)

Material Tx [C ]

GeTe 175

GeTeC 5% 280

GeTeC 10% 320

GeTeC 15% 365

GeTeC 20% 380

GeTeN 5% 235

GeTeN 10% 265

GeTeN 15% 290

GeTeN 20% 310

Table 2.1: Crystallization temperatures Tx of C and N doped GeTe films (150 nm thick),

taken as the midpoint of the rising steps of the reflectivity curves reported in Fig.2.3 [45].

location in the crystalline structure. This is essential for understanding if the

structure of the materials remains stable after many cycles of crystallization and

reamorphization.

In the following, experimental results obtained through X-ray scattering on

the structure of amorphous C-doped and N-doped GeTe will be presented. Ab

initio simulations have also been performed in order to have a better under-

standing of the experimental results. The results presented in this chapter have

been published in Ref. [46].

2.2 Theory of the Pair Distribution Function

(PDF) g(r)

The amorphous phase is characterized by the absence of the long-range periodic

order that can be found in the crystal phase, but a short and medium-range order

is still present (see section 1.4.2). The determination of the Pair Distribution

Function (PDF) g(r) allows to have an insight on these local orders, describing

the correlations between pairs of atoms. In the following, the theory at the basis

of the PDF determination will be described [47, 48].

40


Figure 2.5: Schematic representation of an X-ray incident beam scattered by a point-like

sample. The incident wavevector is k0, the scattered wavevector is kf and the momentum

transfer is Q = k0 − kf .

Let us consider a X-ray incident beam on a point-like sample containing N

atoms as depicted in Fig.2.5. The incident beam is characterized by a wavevector

k0 of modulus 2π/λ and an energy E0. The scattered beam is characterized by

a wavevector kf and an energy Ef . The momentum transfer is Q = k0 − kf

and the energy transfer is hω = E0 −Ef . The double differential cross-section is

defined as the number of photons (or neutrons) scattered per unit of time into

the solid angle interval [Ω,Ω + dΩ] and into the energy interval [Ef ,Ef+dEf ]. For

X-Ray scattering, the incident beam energy is of the order of several keV so that

the maximum energy transfered between the incident photon and the sample is

negligible compared to the incident energy and k0 ≈ kf . As a consequence, the

modulus of Q is simply 4πsinθ/λ, with 2θ being the scattering angle, as shown in

Fig.2.5. In scattering experiments no analysis of the out-coming photon energy

is performed, so that all the photons in the solid angle dΩ are measured whatever

their energy, resulting in a cross-section

dσ

dΩ=

∫ E0

−∞

d2σ

dΩdEf

hdω (2.1)

The general expression of dσdΩ

is

dσ

dΩ=

⟨N∑

i,j

fi (Q) fj (Q) eiQ·rij

⟩

(2.2)

41


where fi (Q) is the atomic form factor of atom i at the position ri and rij = rj−ri.

In the equation, i can be equal to j. The notation 〈〉 in Eq. 2.2 denotes

an average on all possible positions of the scattering centers due to thermal

agitation. In order to understand the meaning of Eq. 2.2, let us first consider

the case of a monoatomic sample where all atoms have the same form factor

f (Q), so that

1

N

dσ

dΩ= f 2 (Q)

⟨

1

N

N∑

i,j

eiQ·rij

⟩

= f 2 (Q) · S (Q) (2.3)

where the so-called structure factor S (Q) is defined as

S (Q) =

⟨

1

N

N∑

i,j

eiQ·rij

⟩

(2.4)

where i can be equal to j. The structure factor S (Q) is a dimensionless quantity

and tends to 1 whenQ tends to∞. In a crystal, due to periodicity, S (Q) consists

of peaks for Q vectors belonging to the reciprocal lattice. In a disordered system

such as a liquid or a glass, the scattering is isotropic and the structure factor

only depends on the modulus of Q. It exhibits broad oscillations as it will be

seen on Fig.2.6. In the isotropic case, S (Q) can be written as

S (Q) =

⟨

1

N

N∑

i,j

sin (Qrij)

Qrij

⟩

= 1 +1

N

N∑

i 6=j

sin (Qrij)

Qrij(2.5)

where rij is the interatomic distance between atoms i and j. Considering that

also f (Q) depends only on the modulus of Q, Eq. 2.4 can be written as

1

N

dσ

dΩ= f 2 (Q) · S (Q) (2.6)

which can be rewritten as

1

N

dσ

dΩ= F (Q) + f 2 (Q) (2.7)

where F (Q) = f 2 (Q) [S (Q)− 1] is the so-called interference function. It is then

possible to describe the structure of the sample in the real space by means of the

pair distribution function g (r), which can be obtained by Fourier transformation

from the structure factor S (Q) as

g (r) = 1 +1

2π2rρ0

∫ ∞

0

Q [S (Q)− 1] sin (Qr) dQ (2.8)

42


where ρ0 =NVis the atomic number density of the sample. The pair distribution

function g (r) tends to zero for r → 0 and tends to 1 for r → ∞. It can be

demonstrated that

g (r) =1

N

⟨∑N

i 6=j δ (r − rij)⟩

4πr2ρ0(2.9)

In Eq. 2.9 the meaning of the function g (r) is clearly stated. It is proportional to

the probability of finding two atoms at a distance r. It should be also underlined

that 4πr2ρ0 · g (r) dr is the average number of atoms between r and r+ dr. It is

possible to obtain the structure factor S (Q) starting from g (r) by means of an

inverse Fourier transformation

S (Q)− 1 =4πρ0Q

∫ ∞

0

r [g (r)− 1] sin (Qr) dr (2.10)

In a polyatomic system constituted of n distinct chemical species denoted α, β,

etc. it is necessary to define partial pair distribution functions gαβ (r) and partial

structure factors Sαβ (Q). In the formalism developed by Faber and Ziman [49]

one defines

gαβ (r) =1

Ncαcβ

⟨∑

iα 6=jβδ(r − riαjβ

)⟩

4πr2ρ0(2.11)

Sαβ (Q) = 1 +1

Ncαcβ

⟨∑

iα 6=jβ

sin(Qriαjβ

)

Qriαjβ

⟩

(2.12)

where cα and cβ are the concentration of atoms for the species α and β so that∑n

α cα = 1, N is the total number of atoms, iα is an atom of species α, iβ is an

atom of species β and ρ0 =NV

as in the monoatomic case. The sum on iα runs

from 1 to Nα = cαN and the sum on jβ runs from 1 to Nβ = cβN . The partial

pair distribution function gαβ (r) is proportional to the probability of finding

an atom of species α at a distance r from one atom of species β. The partial

structure factor and the partial pair distribution function are related to each

other though the relations

Sαβ (Q) = 1 +4πρ0Q

∫ ∞

0

r [gαβ (r)− 1] sin (Qr) dr (2.13)

gαβ (r) = 1 +1

2π2rρ0

∫ ∞

0

Q [Sαβ (Q)− 1] sin (Qr) dQ (2.14)

43


It is possible to generalize Eq. 2.7 obtained in the monoatomic case as

1

N

dσ

dΩ=

∑

α,β

cαcβfα (Q) fβ (Q)Sαβ (Q)−∑

α

cαcβfα (Q) fβ (Q) +∑

α

cαf2α (Q)

=∑

α,β

cαcβfα (Q) fβ (Q)Sαβ (Q) +∑

α

cαf2α (Q)−

[∑

α

cαfα (Q)

]2

(2.15)

By defining⟨f 2⟩=

∑

α

cαf2α (Q)

and

〈f〉 =∑

α

cαfα (Q)

it is then possible to rewrite the cross-section in Eq. 2.15 as

1

N

dσ

dΩ= 〈f〉2 S (Q) +

⟨f 2⟩− 〈f〉2 (2.16)

where the total structure factor S (Q) is defined as the weighted sum of the

partial structure factors Sαβ (Q)

S (Q) =

∑

α,β cαcβfα (Q) fβ (Q)Sαβ (Q)

[∑

α cαfα (Q)]2(2.17)

It is then possible to define a total pair distribution function g (r) by Fourier

transformation in analogy with the monoatomic case (Eq. 2.8)

g (r) = 1 +1

2π2rρ0

∫ ∞

0

Q [S (Q)− 1] sin (Qr) dQ (2.18)

However, due to the Q dependence of the atomic form factors, g (r) has no direct

physical meaning. In particular it cannot be expressed as a weighted sum of the

partial pair distribution functions gαβ (r).

The total structure factor can be deduced from the measured scattered in-

tensity as a function of the scattering angle 2θ , after several experimental cor-

rections (see appendix A.2.2). It is important to underline that through the

experiment it is possible to measure only the total structure factor S (Q) and

not the partial ones. The atomic form factors increase with the electron number

in the atom (in the limit Q = 0 fα (0) = Zα, the number of electrons). From Eq.

44

2.3 Description of the samples

2.17 it can be deduced that, in presence of heavy elements, the contribution of

light elements to the total S (Q) will be relatively small, especially if they are

in low concentration. This will be the case in the following for all partial terms

involving C or N in carbon and nitrogen doped GeTe.

It is indeed possible to compute through ab initio simulations either all the

partial pair distribution functions gαβ (r) or the partial structure factors Sαβ (Q).

The latter can then be summed according to Eq. 2.17 to obtain a calculated

total structure factor Scalc (Q) which can be directly compared to the measured

one Sexp (Q). In order to compare experiments and calculations in r, the Fourier

transform of Scalc (Q) and Sexp (Q) must be calculated according to Eq. 2.18,

leading to gcalc (r) and gexp (r) respectively. However, gcalc (r) cannot be directly

obtained by summing calculated gαβ (r) as already remarked above.

2.3 Description of the samples

Amorphous thin films 200 nm thick of GeTe, C-doped GeTe containing 9.5 and

16.3 at. % of C and N-doped GeTe containing 4 and 10 at. % of N were deposited

by sputtering on a Si (100) substrate as described in appendix B. The Ge and Te

concentrations in the films were measured by Rutherford Back Scattering (RBS)

and the N and C dopant concentrations by Nuclear Reaction Analysis (NRA). In

all the studied samples, there is a small excess of Ge corresponding to Ge52Te48,

but in the following we will refer to the material as GeTe for simplicity. The

atomic number densities deduced from the measured mass densities of all the

samples are reported in table 2.2.

To perform the X-ray scattering experiments, the films have been gently

scratched in order to obtain a powder and the samples have been prepared by

putting about 1 mg of that powder inside a borosilicate glass capillary (700 µm

diameter). The packing fraction was of the order of 10%.

45

2.4 X-Ray scattering measurements and results

Material Mass density [g/cm3] Atomic number density [atoms/A3]

GeTe 5.38 0.0327

GeTeC9.5% 5.16 0.0342

GeTeC16.3% 5.12 0.0363

GeTeN4% 5.285 0.03328

GeTeN10% 5.22 0.0347

Table 2.2: Measured mass densities and atomic number densities for Ge52Te48, undoped and

doped with carbon or nitrogen, expressed in g/cm3 and atoms/A3, respectively. The mass

densities have been measured by X-ray reflectivity (XRR).


The X-Ray scattering experiments were performed on beamline CRISTAL at the

SOLEIL Synchrotron (Saclay, France) at an incident energy of E=45.4793 keV

(λ = 0.4441A) in transmission geometry. The transmitted scattered intensity

was collected by a 2D image plate detector. A photo and a schematic represen-

tation of the experimental setup are shown in appendix A.2.2 in Fig.A.4. The

capillaries filled with powder were installed on a small goniometer which was

rotating during the measurements in order to average possible non uniformity

of the powder in the capillary. For each sample the coefficient of transmission

was determined. The empty capillary signal has been measured to be used

for subtraction. Considering the distance D ≈ 21cm between the sample and

the detector, scattering by air is not negligible. Thus, the air signal has been

determined by measuring the scattered intensity without sample or capillary.

The intensity as a function of 2θ has been obtained for each sample by in-

tegrating the diffracting rings over a vertical sector of the 2D image and by

correcting the obtained intensity as explained in details in appendix A.2.2. The

resulting intensity as a function of 2θ can easily be transformed through the

relation Q = 4πsinθ/λ in a function of Q, I (Q), which is proportional to the

cross section 1N

dσdΩ. The total structure factor S (Q) can be obtained from I (Q)

through Eq. 2.16 after adequate normalization. The pair distribution function

g (r) is then calculated from S (Q) through Fourier transform (see Eq. 2.18)

46


over a range of Q, that in our case has been chosen between 0.1 and 12 A−1,

and by using the densities of Table 2.2.

The S(Q) and g(r) of the undoped GeTe sample are shown in Fig.2.6. They

both corresponds very well to those reported in Ref.[27]. From ab initio simula-

tions, [33] it is known that the first peak of g (r) (at about 2.6 A) is due to the

contributions of mostly Ge-Te bonds with only a few Ge-Ge bonds. The Te-Te

bonds give the principal contribution to the second peak of g (r) (around 4.1

A). In Fig. 2.7 the g(r) of amorphous GeTe is compared to that of crystalline

GeTe measured under the same experimental conditions. In crystalline GeTe,

the contribution to the first peak is due to the short and long Ge-Te distances

in the rhombohedral structure while the second peak is due to Ge-Ge and Te-Te

second neighbours (see Fig.1.11).

In Fig.2.8 the first two peaks of the pair distribution functions obtained for

the GeTeC (a) and GeTeN (b) samples are reported and compared to undoped

GeTe. In all cases, the first peak at around 2.6 A remains unchanged, indicating

that the first distances are not affected by doping, while there is an effect on

the second peak. A new peak appears at around 3.5 A in the doped samples

and the intensity of the peak at around 4.1 A is reduced. If one observes the

differences between the samples doped with different dopants concentrations it

is clearly visible that the effects increase as a function of doping and are more

marked for the GeTeC samples than for the GeTeN samples.

It is important to recall that through the experiment it is possible to measure

only the total functions and not the partial ones, as already underlined in section

2.2. The contribution of the various elements to the factors 〈f〉2 and 〈f 2〉 of

Eq. 2.15 is proportional to the concentration of each element and to its atomic

number Z. Thus, the contributions of bonds involving C or N is negligible due

to the low atomic number and weak concentration of C and N atoms compared

to Ge and Te atoms. For instance, in C-doped GeTe containing 16.3 at. % of

C the contribution of pairs involving C (Ge-C, Te-C and C-C) to the measured

g(r) is much smaller (respectively, 1.6%, 2.8% and 0.04%) than that of Ge-

Ge (12.5%), Ge-Te (44.1%) and Te-Te (38.8%) pairs. Only the Ge-Ge, Ge-Te

and Te-Te bonds contribute to the total pair distribution function determined

47


0 2 4 6 8 10 12 14 16 18 200.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

S(Q

)

Q[Å-1]

2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0-0.5

0.0

0.5

1.0

1.5

2.0

2.5

3.0

(b)

g(r

)

r[Å]

(a)

Figure 2.6: Measured (a) S (Q) and (b) g (r) for undoped amorphous GeTe. It can be noted

that S (Q) tends to 1 for high Q.

48

2.5 Ab initio simulations

2.0 2.5 3.0 3.5 4.0 4.5 5.0-0.5

0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

g(r

)

r[Å]

Figure 2.7: Comparison between the measured g(r) of amorphous (blue) and crystalline

(red) undoped GeTe.

experimentally and one can conclude that the new peak appearing for the doped

samples is not due to bonds directly involving C or N.


Considering that only the total pair distribution functions can be measured ex-

perimentally, it is impossible to deduce the origin of the peak of g(r) that appears

for doped samples. In order to go further in the analysis, ab initio molecular

dynamics simulations were performed for both GeTeC and GeTeN by Dr J-Y.

Raty of the University of Liege. The simulations were done on 210 atom boxes

(92 Ge, 86 Te, 32 C for the C-doped sample and 97 Ge, 92 Te, 21 N for the

N-doped sample) corresponding to Ge52Te48 doped with 15 at. % C and 10 at.

% N. The initial positions of Ge and Te atoms were those found in a previous

simulation of amorphous GeTe [50]. Then C (or N) atoms were introduced by

randomly substituting Ge and Te atoms. The electronic properties were com-

49


2.0 2.5 3.0 3.5 4.0 4.5 5.0-0.5

0.0

0.5

1.0

1.5

2.0

2.5

3.0

g(r) GeTe

g(r) GeTeC 9.6%

g(r) GeTeC 16.3%

g

(r)

r[Å]

2.0 2.5 3.0 3.5 4.0 4.5 5.0-0.5

0.0

0.5

1.0

1.5

2.0

2.5

3.0

g(r) GeTe

g(r) GeTeN 4%

g(r) GeTeN 10%

(b)

g(r

)

r[Å]

(a)

Figure 2.8: Measured g (r) for (a) undoped GeTe and GeTeC (C=9.6% and 16.3%) and (b)

undoped GeTe and GeTeN (N=4% and 10%). In both cases, the first peak is constant with

doping while the intensity of the second peak of the undoped sample decreases with increasing

doping contents. A new peak appears at around 3.5 A in the doped samples. These effects

increase as a function of doping and are stronger in the GeTeN case.

50


puted within density functional theory (DFT) implemented in the VASP code,

using PAW potentials and the PBE-GGA exchange correlation as described in

Ref.[51]. The liquid system was simulated at 3000 K for 3 ps, and then ther-

malized at 1073 K for 10 ps. It was cooled down to the final temperature of

300 K with a 30 K/ps ramp. The density was adjusted to minimize the effective

stress on the simulation box. The refined atomic densities are 0.033 atoms/A3

for undoped GeTe, 0.0343 atoms/A3 for GeTeC15% and 0.0337 atoms/A3 for

GeTeN10%, close to what measured experimentally. The amorphous structural

data were gathered by averaging relevant quantities over a 10 ps trajectory at

300 K.

From the simulation it is possible to obtain all the partial pair distribution

functions gαβ (r) and so it is possible to evaluate the contribution of each atomic

pair, including the ones involving light elements that could not be observed

experimentally. The total stucture factors Scalc (Q) have been obtained from

weighted partial structure factors as described in Eq. 2.17. The pair distribu-

tion functions gcalc (r) have been then calculated from Scalc (Q) though Fourier

transform as described in Eq. 2.18, using the same window 0.1< Q <12A as

for the experimental data, and the results are shown in Fig.2.9. For undoped

GeTe, the calculated gcalc (r) is in good agreement with literature [28, 33, 52].

In the calculated gcalc (r) the first peak remains almost constant with doping,

the second peak decreases with doping and a new peak appears around 3.3A for

the doped samples. The effects of C and N doping on the amorphous structure

of GeTe are comparable.

By comparing the simulated and experimental g (r) in Fig.2.9, a shift in the

peak positions can be observed. The existence of this shift is already known

from literature [53, 54]. Besides, in the experimental g (r) the effect of N doping

is weaker than the one of C doping. However, apart from those two differences,

the good agreement between the evolution as a function of doping for the sim-

ulated and measured total pair distribution functions allows to conclude that

the ab initio simulation takes properly into account the effect of dopants on the

amorphous structure of GeTe.

The next step is to analyze the information that can be obtained from a visual

51


2.0 2.5 3.0 3.5 4.0 4.5 5.0

0

1

2

3

4

5

6

7

8

g(r) GeTe

g(r) GeTeC 16.3%

g(r) GeTeN 10%

Simulation

g(r

)

r[Å]

Experiment

Figure 2.9: Comparison between measured and calculated g (r) for undoped GeTe, GeTeC

(C=16.3% in the experiment and C=15% in the simulation) and GeTeN (N=10% both in the

experiment and in the simulation). Even if an already known shift between peaks positions

can be observed, the evolution of the simulated and measured pair distribution functions with

doping are in good agreement. The effect of N-doping is stronger in the calculated g (r) than

in the measured one.

52


1 2 3 4 5

0

1

2

3

4

5

6

7

8

9

10

11

12

13

1 2 3 4 5

0

5

10

15

20

25

30

35

40

45

50

(b)

(a) GeTe

GeTeC 15%

GeTeN 10%

Ge-Ge

Te-Te

g(r

)

r[Å]

Ge-Te

C-Te/N-Te

C-Ge/N-Ge

C-C/N-N

g(r

)

r[Å]

GeTeC 15%

GeTeN 10%

Figure 2.10: Partial pair distribution functions for (a) Ge-Ge, Te-Te, and Ge-Te pairs in

doped and undoped samples and (b) pairs involving C or N. Curves are shifted for clarity. A

new peak appears in the range 3.1-3.5 A in the Ge-Ge partial pair distribution function for

both C doped and N doped samples, while it is absent in the undoped sample. A difference

between partial contributions involving C and the ones invloving N is the absence of Te-N

bonds at small distances (less than 3.2 A).

53


inspection of the simulation boxes and from the calculated partial pair distribu-

tion functions in order to understand the local atomic arrangements. The partial

pair distribution functions calculated for GeTe, GeTeC15% and GeTeN10% are

shown in Fig.2.10. For simplicity, in Fig.2.10(a) the pair contributions due to

Ge-Ge, Ge-Te, and Te-Te pairs in doped and undoped samples are reported,

shifted for clarity, while the pairs involving C or N are reported in Fig.2.10(b).

From Fig.2.10(a) it is evident that the contributions to the first peak are given by

a few Ge-Ge bonds, at an average distance of 2.63 A for GeTe and GeTeC sam-

ples and 2.65 A for the GeTeN sample, and mostly by Ge-Te bonds at an average

distance of 2.74 A in GeTeC and 2.76 A in GeTe and GeTeN. For both doped

and undoped GeTe the most probable Te-Te distance is rather large (around

4.1 and 4.2 A), with only a few Te-Te bonds about 2.95 A. This corresponds

to a tendency for a strong chemical order due to the alternation of Ge and Te

atoms that is the precursor of the rhombohedral crystalline GeTe structure [55]

and this tendency is maintained in presence of doping. The most relevant result

is the appearance of a new peak at around 3.39 A in the Ge-Ge partial pair

distribution function for both C doped and N doped samples, while it is absent

in the undoped sample.

Let us focus first on the analysis of the C-doped sample. From Fig.2.10(b) the

presence of C-C bonds at a very short distance (1.31 A) can be noticed. These

bonds were not present in the initial state of the system and correspond to a

p-type of bonding. By observing the snapshot of the final state of the simulation

box for GeTeC reported in Fig.2.11, C chains of various lengths (2−5 atoms) can

be observed, without any branching. Observing Fig.2.10(b) it is also possible to

notice many C-Ge bonds that contribute to the partial pair distribution function

at an average distance of 2.05 A. Also a few C-Te bonds are present, with a

contribution at around 2.23 A, but they are much less frequent than C-Ge bonds.

The analysis of bond angle distribution shows that Ge atoms for undoped

GeTe are either in a tetrahedral environment (about 25%-35%) or in a strongly

distorted octahedral environment. In C-doped GeTe, the bond angle distribu-

tion becomes broader making the distinction between tetrahedral and octahedral

environment less clear. The inspection of the model structure (Fig.2.11) com-

54


Figure 2.11: Snapshot of the final state of the simulation box for GeTeC. Ge atoms are

represented in pink, Te atoms in light blue and C atoms in red. The inspection of this box

combined with a bond angle analysis around C atoms, reveals the presence of a mixture of

tetrahedral (C−TeGe3, C−Ge4 and C−Ge2Te2), triangular (C−C−Ge2 and C−C−GeTe),

and linear (C chains) bonds.

55


Figure 2.12: Summary of the carbon environments in the C-doped GeTe sample. C−TeGe3,

C−Ge4 and C−Ge2Te2 tetrahedra can be found, as well as C−C−Ge2 and C−C−GeTe

triangular environments.

bined with a bond angle analysis around C atoms reveals the presence of a

mixture of tetrahedral (sp3), triangular (sp2), and linear (sp) C-Ge bonds. The

C−TeGe3 tetrahedra are the majority (3 are present) and one C−Ge4 and one

C − Ge2Te2 tetrahedra can also be found, in agreement with the relative ratio

of C-Ge and C-Te bonds. Eleven C−C−Ge2 and one C−C−GeTe triangular

environments are also present. A schematic representation of the environments

involving carbon is depicted in Fig.2.12. Two Ge atoms belonging to the same

tetrahedral unit centered on a carbon atom result to be separated on average

by 3.39 A and that is the same distance for two Ge of a C − C − Ge2 triangle.

This Ge-Ge distance is induced by the presence of C and it can be clearly seen

in Fig.2.10(a), where a new peak appears at 3.39 A in the Ge-Ge partial pair

distribution of GeTeC. In the r-range from 3.2 to 3.6 A the Ge-Te and Te-Te

partial pair distribution are only slightly affected by C doping.

Let us now analyze the N-doped sample. By comparing with GeTeC, the

first striking difference that can be noticed is the absence of Te-N bonds at

small distances (less than 3.2 A) for GeTeN while there are a few Te-C bonds

56


Figure 2.13: Snapshot of the final state of the simulation box for GeTeN. Ge atoms are repre-

sented in pink, Te atoms in light blue and N atoms in green. N−Ge3 pyramidal environments,

N−Ge4 tetrahedral environments and N2 molecules can be found.

57


Figure 2.14: Summary of the nitrogen environments found in the N-doped GeTe sample.

N−Ge4 tetrahedra, N−Ge3 pyramids and N2 molecules have been observed.

at around 2.23 A in GeTeC. The only bonds involving N that can be found

in GeTeN for r < 3.2A are Ge-N and N-N bonds. In the snapshot of the

final state of the simulation box for GeTeN reported in Fig.2.13, 8 N − Ge3

pyramidal environments, 9 N − Ge4 tetrahedral environments and 2 molecules

of N2 can be found. The observed environments for N-doped GeTe are reported

in Fig.2.14. It should be noticed that during the simulated annealing procedure,

the initial configuration was taken from a liquid sample equilibrated at a very

high temperature (3000 K), so that any kind of possible bond was present in the

structure, including Te-N bonds. However, due to the low N concentration, no

N2 molecule was present in the liquid. It is thus striking to find that all the Te-N

bonds have been disrupted upon cooling and statistically unlikely N2 molecules

have formed.

58

2.6 Discussion

2.6 Discussion

The two main conclusions of this work are that, on one hand, the strong chemical

order between Ge and Te characteristic of amorphous GeTe is maintained in

presence of C or N doping elements and that, on the other hand, dopants deeply

modify the structure of the amorphous phase by introducing new environments.

A common feature of the C and N doping is the appearance of tetrahedral

units centered on C or N. In C-doped amorphous GeTe, triangular environments

around C and short C chains (containing between 2 and 5 C atoms) can also be

found, while in N-doped GeTe NGe3 pyramids and N2 molecules are also present.

One major difference between N and C doping is the fact that C can form short

bonds with Te, although in smaller proportion than C-Ge and C-C bonds, while

short N-Te bonds are absent in N-doped GeTe. The same conclusion has been

recently obtained in Ref. [56] through X-ray absorption spectroscopy and X-ray

photoemission spectroscopy (XPS). The fact that C is bonded preferably to Ge

has also been found by ab initio simulations in the case of C-doped GST in

studies performed at the same time as the study reported in this thesis [57, 58].

In N-doped GST, ab initio simulations show that N is bonded in majority to

Ge, to Sb in a less extent and marginally to Te [59].

In C-doped GeTe, two Ge atoms belonging to the same tetrahedral or tri-

angular unit are separated by 3.4 A. This Ge-Ge distance, that did not exist

in undoped amorphous GeTe, appears as a new peak in the Ge-Ge partial pair

distribution function and is also visible in the calculated total g (r). From these

ab initio simulations, it can be concluded that the additional peak observed

around 3.3 A in the measured pair distribution function of GeTeC (C=9.5%)

and GeTeC (C=16.3%) is mainly due to this new Ge-Ge distance present in

tetrahedral and triangular units centered on carbon. The contribution of Ge-Te

and Te-Te distances to the total g (r) in this r-range is much less than that of

Ge-Ge distances because of the small number of C-Te bonds compared to C-Ge

bonds. In N-doped GeTe two Ge atoms belonging to the same tetrahedral or

pyramidal unit are separated by 3.25 A. They generate a new peak in the Ge-Ge

partial pair distribution function and in the calculated total g (r).The effect on

59

2.6 Discussion

the measured g (r) of GeTeN (N=10%) is smaller than predicted by the calcu-

lation. One explanation could be that the sample contains more N2 molecules

than the simulation box so that the relative abundance of NGe4 and NGe3 units

is reduced in the sample. Information from literature on the presence of N2

molecules in N-doped amorphous GeTe is contradictory at the present time. N2

molecules have been found by X-ray absorption and XPS experiments in Ref.[60]

but not in [56].

It is worth noting that the samples studied experimentally in this work are

as-deposited amorphous samples, while the simulation results are collected on

an amorphous state obtained by a rapid quench of the liquid phase (the so

called melt-quenched amorphous). The structure of these two amorphous states

could differ. This question has been addressed in literature in the case of N-

doped GeTe by ab initio simulations [54]. The conclusion of this study is that

the difference in the local structure between as deposited and melt-quenched

amorphous phase is only the number of tetrahedral environments.

One can wonder whether the structural changes observed in doped GeTe can

be related to the enhanced crystallization temperature and activation energy of

C and N-doped GeTe. The presence of C or N makes the Ge environments, on

average, slightly more tetrahedral. According to the ideas developed in Ref.[50],

this structural change would increase the average number of constraints so dop-

ing could indeed increase the stability of the amorphous state. This idea led

to a study of the vibrational properties of C and N-doped amorphous GeTe

by combining Fourier Transform Infrared spectroscopy (FTIR) experiments and

computation of the vibration modes from the ab initio simulations presented in

section 2.5. The main conclusion of this study (Ref.[61], submitted for publica-

tion) is that the inclusion of C and N creates high frequency, localized modes,

and at the same time decreases the density of low frequency acoustic-like modes.

In Ref.[61] it is shown that this kind of effects corresponds to an increased sta-

bility of doped amorphous GeTe. Besides, in the case of C-doped GeTe the

presence of C chains could be an obstacle for the crystallization of the GeTe

matrix, provided that these C-C bonds disappear in the crystalline phase, since

breaking a C-C bond is an extremely endothermic process.

60

2.7 Conclusions and perspectives


In conclusion, the addition of carbon or nitrogen in GeTe has been observed to

deeply modify the structure of the amorphous phase. Both dopants induce the

formation of new environments and bonds, resulting in a more stable amorphous

structure , which could explain the increase of the crystallization temperature

and activation energy induced by doping. In parallel to the work described in

this chapter, the effect of carbon doping on the properties of GST has been

observed experimentally very recently [62]. In that case ab initio simulations

[57, 58] have preceded the experimental study. The conclusion of all these works

is that carbon doping enhances the stability of the amorphous GST phase, like

in the case of C-doped amorphous GeTe studied in this thesis.

An enhanced stability of the amorphous phase would produce an increase of

the retention time, beneficial for PCM. Such an increase is indeed observed in

PCM containing C on N doped GeTe [44, 41]. However, understanding its origin

requires further studies since other effects could contribute as well. From the

device point of view, the capability of the material to maintain its properties after

many phase transformations is an imperative requirement. Therefore, starting

from the understanding of the amorphous structure of C and N doped amorphous

GeTe, the next step must be the study of the crystalline phase. At the moment it

is unclear if the doping elements remain in the crystalline phase, at substitution

or interstitial sites, or if they segregate, for instance at the grain boundaries. In

the latter case one can wonder if they are incorporated again in the amorphous

phase upon reamorphization.

Little is known on the crystalline structure of C and N doped GeTe. One

result is that the crystalline phase is cubic at room temperature for C and N

content above 10% [60, 63]. In the case of crystalline N-doped GST (NGST) dis-

tortions of the crystal lattice and reduction of the grain size have been observed

[34, 64]. The conclusion was that the N atoms occupy tetrahedral interstitial

sites in the NGST crystalline structure and that a Ge3N4 phase is formed near

the grain boundaries, thus reducing the grain size. Moreover, the presence of

some N2 molecules was also observed. The molecular nitrogen is believed to

61


exist at interstitial and vacancy sites, and more likely at grain boundaries [37].

Further studies are required in order to localize the dopants in the crystalline

phase. Characterization techniques such as X-ray photoemission spectroscopy

and Fourier transform infrared spectroscopy will be useful to provide a detailed

comparison of the amorphous and crystalline phases and to investigate the evolu-

tion between them. Finally, local analysis via transmission electron microscopy

could be useful to study amorphous and crystalline phase in both thin films and

integrated materials.

62

Chapter 3

Confinement of phase change

materials: Ge2Sb2Te5 nanoclusters

63

3.1 Introduction

3.1 Introduction

One of the open questions about the competitiveness of PCM is at which extent

their dimensions can be reduced. Their theoretically very high scalability is one

of their most promising properties, as it can make them become competitive with

Flash memories. In terms of devices, the scaling of the memory cell is limited by

architectural and geometrical issues and by material scaling possibilities. Thus,

it is important to investigate the effect of shrinking size on the phase change

materials properties and characteristics. First, it is of the uttermost importance

to establish if there is a limit size under which the phase transformation can no

longer occur. Moreover, as it will be discussed further in the following sections,

the effect of reduced dimensions on the mechanisms of crystallization is still an

open question. This confinement effect can be studied at different extents: the

first approach is to investigate the effect of reducing thickness of PC thin films

on the phase change mechanism. A further step consists in investigating con-

fined nanostructures of PC materials, such as nanowires and nanoclusters. In

this chapter, an overview of the results obtained on confined structures will be

presented. A new method for the deposition of PC nanoparticles of diameter

below 10 nm will be introduced and results obtained on Ge2Sb2Te5 (GST) parti-

cles embedded in Al2O3 will be presented. The results presented in this chapter

have been published in Ref. [65].

3.1.1 Effect of shrinking size in one dimension: thin films

Many studies regarding the scalability of PCM deal with the thickness depen-

dence of the amorphous to crystalline phase transition in PC thin films, meaning

the effect of confining the material in only one dimension. In the following, a

review of the main results from literature on GST will be presented. First, evi-

dences of the influence of thin films thickness on the crystallization temperature

Tx will be shown, then the crystallization mechanisms will be discussed and at

the end some interesting results on multilayers will be reported.

In literature, the crystallization behavior of a PC material has been inves-

tigated by several experimental methods, as for example optical measurements

64

3.1 Introduction

[66], electrical measurements [67], Transmission Electron Microscopy (TEM)

[68], Differential Scanning Calorimetry (DSC) [69] or X-Ray Diffraction (XRD)

[70]. All the quantities measured by these methods are directly proportional to

the quantity of crystalline matter in the sample, except the case of resistivity

measurements where the value of resistivity decreases only when a crystalline

path between electrodes is formed. There is no generally admitted definition of

the crystallization temperature Tx and different criteria are used. For example,

it can be defined as the temperature at which the first derivative of a reflectivity

curve with respect to temperature has a maximum value, or as the point at

which the resistance of a sample is reduced by a factor of 2, or as the peak point

in a Differential Scanning Calorimetry experiment. The experimental conditions

are different for all these techniques, and considering that crystallization is ruled

by kinetic theories, the value of Tx depends on the thermal history of the sample

and thus on these conditions. This means that the value of Tx depends on both

the experimental conditions and on the criteria used to define it.That said, under

the same experimental conditions, it is widely assumed that a PC material film

at least 100 nm thick should be characterized by a crystallization temperature

Tx that is independent form the surrounding layers and can be considered as its

bulk crystallization temperature.

In 2007, Wei et al. measured the amorphous to fcc rocksalt phase transition

temperature, the crystallization speed and activation energy (Kissinger method)

for thin films of GST of various thicknesses (5, 10, 15, 20 and 30 nm) sandwiched

by 50 nm of ZnS − SiO2 [67]. The crystallization temperature Tx increases

as the film thickness d is reduced below 20 nm, as can be seen in Table 3.1,

while the crystallization speed decreases (as shown in Figure 3.1). No thickness

dependence of parameters has been found for films thicker than 20 nm. The

estimated crystallization activation energy increases with decreasing thickness,

from 2.86 eV for 20 nm thick films to 4.66 eV for 5 nm thick films, and a linear

relationship has been found between ln[Tx] and the thickness d of the GST thin

films. Moreover, the Avrami coefficient n (described in section 1.3.5) has been

measured and it has been found that n > 1.5 for d ≥ 20 nm, while 1 < n < 1.5

for d = 10 nm and n < 1 for d = 5 nm. This suggests that the nucleation and

65

3.1 Introduction

Figure 3.1: Resistivity as a function of time at room temperature for thin films of GST of

different thicknesses pre-annealed at 143.5C (from Ref.[67]). The incubation time τ , defined

as the time elapsed before the onset of crystallization, and the transition time from the highest

to lowest resistivity increase with decreasing film thickness, meaning that the crystallization

speed is reduced for small thicknesses.

Figure 3.2: Model used Ref.[71] to interpret the thickness-dependent variation of Tx . A

cylindric crystalline nucleus is embedded in the amorphous phase, sandwiched between two

oxide interfaces.

66

3.1 Introduction

Heating rate (C /min) 5 nm 10 nm 15 nm 20 nm 30 nm

0.5 157 149 141 138 138

1 159 153 147 142 142

3 164 157 152 148 148

10 166 160 155 154 154

20 170 164 160 157 157

Table 3.1: Crystallization temperatures as a function of heating rate and GST film thickness

from Ref.[67].

growth mechanisms change by decreasing thickness.

In order to explain the thickness-dependent increase of Tx for thin films of Si

with oxide interfaces a model has been proposed by Zacharias [71] and it has been

be applied to PC materials. In this model, based on homogeneous nucleation,

an effective interfacial energy between the oxide capping layer and the surface

of a growing crystalline cylindrically shaped nucleus in the amorphous matrix

(see Fig.3.2) is introduced. This effective interfacial energy is a function of the

distance between the nucleus surface and the oxide. By calculating the Gibbs

free energy for the nucleus, the crystallization temperature Tx can be expressed

as an exponential function of the phase change film thickness d

Tx = Tac

[

1 +

(γoc − γac − γoa

γace−d/4l0

)]

(3.1)

where γac, γoc, and γoa are defined as the interfacial free energies per unit area

between the amorphous and crystalline phases (ac), between the oxide capping

layer and the crystalline phase (oc), and between the oxide capping layer and the

amorphous phase (oa), respectively. Tac is the bulk crystallization temperature

and l0 can be interpreted as an average screening or bonding length which is

related to the range of interatomic forces typical for the materials o and c. The

values reported in Table 3.1 can be fitted by Eq. 3.1, as reported in Ref.[67].

Another study on the effect of reducing thin film thickness over the crystal-

lization temperature Tx has been published by Raoux and coworkers [70]. The

value of Tx has been measured for films of various thicknesses (from 1 to 50

nm) and different materials (GST, N-doped GST, Ge15Sb85, Sb2Te and Ag- and

67

3.1 Introduction

In-doped Sb2Te). All the samples were deposited on Si and capped with 10

nm of Al2O3 . They have been characterized through X-Ray diffraction during

in-situ annealing with a heating rate of 1C /s. The transition temperature Tx ,

corresponding to the amorphous to fcc rocksalt structure transformation, is sim-

ilar to that of the bulk (close to 155C ) for all films with thicknesses above 10

nm. Tx increases with decreasing thickness below 10 nm for all the considered

materials, but at a different extent for each composition. The value reported

for Tx is 342C for a 5 nm thin GST film and 380C for a 2 nm thin GST film,

where the phase transition proceeds directly from the amorphous to the hexag-

onal phase. It can be noted that the variation of Tx with thickness for the GST

films capped with Al2O3 of Ref. [70] is much higher than for the GST films of

Ref. [67] capped with ZnS−SiO2. It is also worth noting that the so called bulk

crystallization temperature is 155C in Ref.[70] and 142C in Ref.[67] for the

same heating rate. This could come from the different experimental methods

(resistance measurements in Ref.[67] and X-ray diffraction in Ref.[70]). Other

possibilities are that the composition of the material is not exactly the same

(there can also be an inclusion of undesired doping elements) or that the dif-

ferent interface materials still influence Tx in films of 30-50 nm thickness (see

Chapter 4). The thickness dependence of Tx in Ref. [70] have been interpolated

using Eq. 3.1 [71] as it is shown in Fig.3.3. On the other side, the transition

from the cubic phase to the hexagonal phase occurs at the same temperature for

all film thicknesses, around 450C . The thinnest GST film that shows evidence

of Bragg peaks is 2 nm thick, so the authors of Ref. [70] conclude that this

thickness is the size limit for the existence of a phase change transformation for

GST.

In 2009, Simpson et al. measured the crystallization temperature of thin

films of GST of thicknesses between 2 nm and 10 nm encapsulated by TiN or

ZnS− SiO2 [72]. In the case of TiN an increase of Tx with decreasing thickness

has been obtained, similar to the trend observed in Ref.[70] for GST encapsu-

lated in Al2O3 , while for encapsulation with ZnS− SiO2 Tx is decreasing with

thickness. The latter result is in contradiction with what observed in Ref. [67]

and reported in Table 3.1. Both in Refs. [72] and [67] the variation of Tx with

68

3.1 Introduction

Figure 3.3: Crystallization temperature Tx as a function of film thickness for various PC

materials: GST, N-doped GST (NGST), Ge15Sb85 (GeSb), Sb2Te and Ag- and In-doped Sb2Te

(AIST) deposited on Si and capped with Al2O3 , fitted using Eq.3.1, as presented in Ref.[70].

69

3.1 Introduction

decreasing thickness for GST interfaced with ZnS − SiO2 is small, but in Ref.

[72] Tx increases with decreasing thickness while the opposite trend is observed

in Ref. [67]. The authors of Ref. [72] proposed the existence of a correlation be-

tween the mechanical stresses induced by the cladding material and the change

in Tx . One difficulty with this interpretation is that, even in the case where

the stress in the cladding layer is known, the evaluation of the level of strain

induced in the embedded layer is delicate [73]. Other recent studies indicate that

compressive and tensile stress inhibit the rocksalt to hexagonal phase transition

when the GST film thickness is thinner than 10 nm [74].

From all these results it is evident that the effect of confinement on phase

change cannot be separated from the effect of interaction between the interface

material and the PC material itself. The influence of interface materials on crys-

tallization of an amorphous compound had been studied also out of the PCM

context. For example, in 1969 Oki et al. reported that a 10− 30 nm film of Ge

crystallizes at different temperatures if interfaced with a metal [75]. In a study

on Pb/Ge multilayer samples in 1987 [76] a decrease of Tx with decreasing Ge

thin films thickness has been observed, and it has been suggested that the crys-

tallization is interfacially initiated and strongly affected by the Pb layer texture.

In late 1990s, the effect of the interface layer on crystallization mechanism of PC

materials has been extensively studied by Ohshima [66, 77]. The aim of those

studies was to understand the effect of the interface dielectrics commonly used

as capping layers for optical devices due to their optical good properties. Those

kinds of dielectrics include SiO2, Si3N4, Ta2O5, ZnS and ZnS − SiO2. The PC

materials are thin films of 30 nm with variable composition in the Sb2Te3−GeTe

pseudobinary system (named Ge−Sb−Te materials). The results indicate that

the crystallization temperature, the crystallization activation energy and the

nucleation process are affected by the different dielectric interfaces. The effect

of the dielectric films is to accelerate the nucleation in Si3N4 and Ta2O5 capped

samples, to inhibit the nucleation in SiO2 capped sample, and to generate nuclei

even during the grain growth process in ZnS and ZnS − SiO2 capped samples.

The author suggests that these variations may depend on the surface reactiv-

ity and chemical affinity of the film materials [66]. This conclusion has been

70

3.1 Introduction

supported by a further study on Ge − Sb − Te film interfaced with Si3N4 or

ZnS − SiO2, where a difference in the crystalline structure and in the grain

growth process has been evidenced between the two different samples [77]. The

films interfaced with Si3N4 crystallize in a fcc structure and the grain grow grad-

ually until the fcc to hexagonal transition, while for the films interfaced with

ZnS− SiO2 the structure was a mixture of fcc and hexagonal structure and the

grains grew abruptly at around 250C .

In 2009, Jang et al. investigated GST/SiO2 multilayered films with layer

thicknesses of [10.42 nm/10.42 nm]×5 (named M10) and [5.93 nm/5.64 nm]×10

(named M5) [78]. Sheet resistance measurements show that the phase-change

characteristics are affected by the bilayer thickness, as can be seen in Fig.3.4.

In particular, the crystallization temperature seems to be slightly higher for

multilayer films than for a single-layered GST thin film of 100 nm. Compared to

the single-layered GST thin film, the temperature region between the amorphous

to fcc transition and the fcc to hexagonal transition is reduced and it decreases

further as the bilayer thickness of the multilayered structure decreases. A similar

reduction was observed in Ref.[70] for GST with Al2O3 capping, but in that case

the amorphous to fcc transition temperature increased while the fcc to hexagonal

transition temperature remained unchanged, while in Ref.[78] the amorphous

to fcc transition temperature increases only slightly and the fcc to hexagonal

transition occurs at lower temperatures. The difference between the two studies

is again the capping material, that can have a different influence on the two

transitions. Moreover, in Ref.[78] it is reported (without quantitative data) that

the GST film is highly strained for thin multilayers due to the different thermal

expansion coefficients of GST and SiO2.

In conclusion, results from literature on the variation of Tx with thickness

are often unclear and ambiguous, and sometimes even contradictory. The inter-

face materials play certainly an important role in determining the amorphous to

fcc and the fcc to hexagonal transition temperatures, and can maybe have dif-

ferent influences over the two of them. Another important issue to be taken into

account is the possibility of interdiffusion from the capping layer through the PC

material, resulting in a doping effect instead of an interface effect. Size reduction

71

3.1 Introduction

Figure 3.4: Sheet resistance of multilayered films of GST/SiO2 as a function of annealing

temperature, as reported in Ref.[78]. The label M25, M10 and M5 indicate different bilayer

thicknesses (M5 corresponds to the thinnest sample, M25 the thickest one). The dotted lines

correspond to ex situ annealing temperatures used for further analysis in Ref.[78].

effects have been reported for thin films but while in some cases the variation of

Tx is extremely high, in other cases it is very small and not even consistent in

different publications (as for the ZnS−SiO2 interface in Ref.[72] and [67], as dis-

cussed above). The greatest variations of Tx have been reported for Al2O3 [70]

and TiN [72] interfaces, but there are no further studies in literature for those

materials. Some attempts to identify a relation between surface reactivity and

chemical affinity of films and a change in the crystallization mechanisms with

different interface materials have been done [66]. However, the physical and

chemical influence of different capping layers on the phase change mechanism it

is still unknown.

3.1.2 Effect of shrinking size in two and three dimen-

sions: nanostructures

As it was discussed in the previous section, reducing the thickness of PC material

thin films can have a great impact on crystallization mechanism. This has been

observed through the variation of some important phase change properties such

as crystallization temperature or activation energy, which depend also strongly

on the interface materials. However, the studies reported in the previous section

deal only with the effect of confining the PC materials in one dimension. To go

72

3.1 Introduction

Figure 3.5: Scanning Electron Microscopy (SEM) image of as-grown GST nanowires from

Ref.[81].

further, systems of nanostructures can be investigated. For example, nanowires

with length much higher than their diameter can be used to study the effect of

confinement in two dimensions.

In 2006, GeTe and Sb2Te3 nanowires have been obtained through vapor-

liquid-solid (VLS) technique [79, 80]. A Scanning Electron Microscopy (SEM)

image of as-deposited GST nanowires taken from Ref.[81] is reported in Figure

3.5. Interesting results about scaling effects on physical parameters have been

obtained with GST nanowires of different thicknesses, from 20 to 190 nm, fabri-

cated through VLS process [82] on a SiO2 /Si substrate with no capping. This

means that the influence of the interface layers that has such a great impact in

the thin film case is reduced for such nanowires. Two Pt electrodes are directly

written onto the crystalline as-deposited nanowires which can then undergo the

SET and RESET operations many times. So nanowires are not only an impor-

tant system for investigating crystallization at a nanoscale size, but they can

also be directly employed as memory devices, even if a new device architec-

ture needs to be developed [81]. Electrical characterizations of reamorphized

nanowires show a decrease of the recrystallization time at fixed temperature,

an increase in the nucleation rate and a decrease in the activation energy for

decreasing diameter (Figures 3.6a, b and c, respectively).

73

3.1 Introduction

Figure 3.6: Measured values for the (a) recrystallization time at fixed temperature, (b)

nucleation rate and (c) activation energy as a function of nanowires diameter as reported in

Ref.[82].

74

3.1 Introduction

The high surface-to-volume ratio of nanowires seems to lead to enhanced

heterogeneous nucleation that can be responsible for the observed size-depending

effects. These results are in contrast with the increase of the crystallization

temperature with decreasing thickness reported for thin films in section 3.1.1,

but nanowires differ significantly from thin films from many points of view.

Their surface-to volume ratio is different and nanowires of Ref.[81] are uncapped

while a capping layer is usually deposited for thin films. Moreover, thin films

are amorphous as-deposited and it is not possible to reamorphize them after

crystallization. So in the amorphous to crystalline transition for thin films the

starting point can be only an as-deposited amorphous phase. For nanowires,

on the other side, crystallization of a melt-quenched amorphous phase has been

studied, so it may be hazardous to directly compare thin films and nanowires

results.

It is possible to go even further in investigating the effects of confinement

on phase change material by studying PC clusters. Due to the 3D confinement,

clusters are the ideal model for investigating the size effects in memory cells

where both size and interface effects play a role. Indeed, in clusters the interface

effects are enhanced due to the large surface/volume ratio. Moreover, for isolated

clusters in a matrix, the plastic relaxation may be limited by confinement, which

can thus enhance strain effects.

Some attempts to obtain PC nanoclusters have been reported [83, 84, 85,

86, 87, 88, 89, 90]. The techniques used to fabricate the clusters vary from

electron beam lithography to laser ablation and chemical synthesis. For GST

nanoclusters made by electron beam lithography with size range from 20 nm

to 80 nm, no significant change in the amorphous to fcc phase crystallization

temperature has been observed [83]. Smaller clusters (15 nm average diameter)

obtained with diblock copolymer transform directly into the hexagonal phase

[84]. The growth of GST clusters using laser ablation has been reported, with

discrepancies in the reported crystallization behavior. The first results [85] with

large size distribution (from 4 nm to 30 nm) indicate that, for 15 nm size-

selected clusters, amorphous particles with irregular shape are obtained when

the annealing is below 200C , while crystalline particles are observed when the

75

3.1 Introduction

annealing is above 300C . One peculiar observation is the fact that, contrary

to bulk or thin films, the hexagonal and fcc phases are formed when annealing is

performed at 300C whereas the pure fcc phase is only observed when annealing

is performed above 400C . A second report, with similar size distribution,

indicates clusters with a mixture of amorphous and fcc phase for all temperatures

from 100C up to 500C [86]. Using a similar growth technique, a third report

indicates mostly amorphous as-prepared particles transforming into a mixture

of hexagonal and fcc phase after annealing at 100C , and in a mostly fcc phase

after annealing at 200C [87].

The main conclusion from these studies is that both the cubic fcc and amor-

phous phases can be observed in nanometric GST clusters. However, the obser-

vation of an unambiguous phase transition from an amorphous to a crystalline

structure, at a definite temperature, has not yet been achieved. Moreover, in

these studies the effect of the cladding material on the phase change proper-

ties of clusters has not been addressed. In order to investigate these aspects,

GST clusters with size below 10 nm and small size distribution are needed.

In the case of GeTe nice results have been obtained using chemical synthesis

[88, 89, 90]. Small particles with size ranging from 1.75 to 3 nm show a Tx more

than 150C above that of the bulk, [89] while the increase for 8.7 nm clusters is

only 67C [90]. However, results are difficult to compare to thin films measure-

ments due to the different fabrication methods and to the possibility of carbon

contamination of the nanoparticles which could contribute to the observed in-

crease of Tx (see Chapter 2). The chemical synthesis of GST clusters has not

been reported up to now. No consistent information have been obtained up to

now on the crystallization temperature of PC nanoparticles beyond 10 nm of

diameter synthesized by sputtering, which is the deposition method commonly

used for thin films deposition.

In conclusion, information in literature about size effect on PC materials

rely on studies of thin films, nanowires and clusters. Thin films show a strong

dependence of the crystallization mechanism on the interface materials and often

exhibit an increase of Tx with decreasing thickness, while nanowires free from

capping materials show a decreasing Tx with decreasing diameter. However,

76

3.2 Clusters deposition

as explained above, it is difficult to interpret the differences observed between

films and nanowires without further studies, for example on capped nanowires.

Nanoparticles are characterized by a high surface to volume ratio and by a strong

influence of capping layers. In the following section the use of a new method

for the growth of GST nanometric clusters by sputtering (section 3.2) and their

characterization through X-ray diffraction (section 3.3) will be described.


In this section the fabrication and deposition of GST nanometric clusters with

average size below 10 nm, embedded in alumina, is described. The process has

been done by R. Morel and A. Brenac (INAC/SP2M, CEA Grenoble) [91]. The

clusters are prepared in a UHV chamber equipped with a sputtering-gas phase

condensation cluster source and two additional standard sputtering guns. The

deposition apparatus is depicted in Fig.3.7. The cluster source itself consists in

a magnetron sputtering head installed in a liquid nitrogen cooled insert. The

stoichiometric Ge2Sb2Te5 solid target is DC sputtered in a 0.1 mbar cold argon

atmosphere, which makes the sputtered vapor condensate into nanometer-sized

clusters. The clusters drift along the gas flow lines and are expelled through

an iris diaphragm in the vacuum, forming a beam which is directed onto the

sample in a deposition chamber. This chamber is equipped with an additional

magnetron that is used for the sputter deposition of GST thin films (using the

same target as the one used for clusters), and a second one for the deposition

of Al2O3 underlayers and capping layers. All depositions are made at room

temperature. The clusters size distribution, given by a time of flight mass spec-

trometer (TOF) , is shown in Figure 3.8a. The average cluster size is 5.7 nm,

and the width of the distribution is ± 1 nm at half maximum. The morphology

of the clusters was controlled by Transmission Electron Microscopy (TEM): a

low density layer of clusters was deposited on an ultra-thin carbon grid, and pro-

tected by a 1 nm Al2O3 layer. The TEM image is reported in Figure 3.8b. The

distribution of the clusters on the surface is random, as expected from the de-

position technique. The particles are spherical. The fact that no atomic planes

77


Figure 3.7: Schematic drawing of the apparatus used to deposit the nanocluster samples

as long as the thin film samples used for comparison. The deposition procedure is briefly

illustrated.

78


Figure 3.8: (a) TOF size distribution which shows that the nanoclusters have an average

diameter of 5.7 nm with a narrow size distribution (± 1 nm at half maximum) (b) TEM images

of GST as-deposited clusters which indicate that the particles are spherical and amorphous.

These images have been made on a low density dedicated sample and the clusters state and

shape have been checked over clusters with the largest diameter. (By courtesy of M. Audier,

LMGP CNRS, Grenoble INP-Minatec

79


Figure 3.9: Scanning Electron Microscopy (SEM) image of GST deposited clusters. The red

circle has a diameter of 20 nm. No trace of particles coalescence can be seen.

are visible for the clusters and that the contrast is similar for all particles is a

first indication that the as-deposited clusters are amorphous, as will be demon-

strated with the X-ray diffraction analysis. In Figure 3.9 a Scanning Electron

Microscopy (SEM) image of the same sample shows no sign of coalescence of the

deposited particles.

Two types of GST samples have been prepared in order to perform the X-ray

diffraction (XRD) measurements. The GST clusters samples consist in 4 GST

clusters layers deposited on Si. First a layer of 6 nm of Al2O3 has been deposited

on the Si substrate. Then a layer of GST clusters with an equivalent mass of

0.07 monolayer of particles has been deposited, covered with a 3 nm Al2O3 layer.

Then, three other layers of GST clusters have been deposited, each one covered

by 3 nm of Al2O3 , and at the end a 6 nm Al2O3 layer has been deposited. The

final structure, shown in Fig. 3.10a, is 6 nm Al2O3 / (GST clusters layer / 3

nm Al2O3 ) × 4 / 6 nm Al2O3 . The average distance between clusters is two

cluster diameters. This structure has been chosen in order to avoid sintering

effects during annealing. The GST film samples consist in 10 nm GST thin film

sandwiched between two 10 nm thin Al2O3 films, deposited on a Si substrate.

80

3.3 X-Ray Diffraction study

Figure 3.10: (a) GST clusters sample: 4 layers of clusters capped with Al2O3 are deposited

on an Al2O3 substrate and capped again with Al2O3 . The average distance between clusters

in a layer is about 2 cluster diameters. (b) GST film sample: 10 nm thick film of GST

sandwiched between two 10 nm thin Al2O3 films. Both the clusters and film samples are

deposited on a substrate of Si.

Their schematic representation is reported in Figure 3.10b. The GST quantity

in clusters samples is five times less than the one in film samples. A 20 nm

Al2O3 thin film on a Si substrate, hereafter called blank sample, was also grown

for background signal measurement in the X-ray diffraction experiments. Thin

films and clusters compositions were measured with Rutherford Backscattering

Spectroscopy (RBS) and Particle-Induced X-ray Emission (PIXE). For thin films

the content is Ge:Sb:Te = 23:24:53 (±3), very close to the Ge2Sb2Te5 = 22:22:56

target composition. For the clusters the composition is 28:27:45 (±3), which

indicates a slight Te depletion as often reported for very thin films and clusters

[70, 83, 86, 92]. Within experimental accuracy, the GST clusters composition is

found identical before and after annealing.


In order to qualitatively locate Tx and observe the crystallized state in clus-

ters a first set of clusters samples was annealed under vacuum (10( − 3) bar) at

200C and 300C , while a film sample was annealed at 200C . The tempera-

81


ture was ramped at + 10C /min, held at the set point temperature for 30 min,

and ramped down to ambient temperature. X-ray diffraction on those samples,

hereafter referred to as ex situ samples, has been performed at room tempera-

ture. For a second set of clusters and thin films samples the X-ray diffraction

spectra were recorded as a function of the temperature, during the annealing

process. Those samples will be referred to as in situ samples. X-ray diffraction

measurements were performed using synchrotron radiation on the BM02 CRG-

D2AM beamline (ESRF Grenoble, France) with a photon incident energy of 17.8

keV (λ =0.69654 A) and using a 2D CCD camera detector. The detailed de-

scription of the experimental setup is reported in appendix A.2.2. The incident

angle was 4 and the position of the CCD camera allowed for measurements in a

2θ range from 8 to 26. As it is reported in details in appendix A.2.2 about the

data treatment description, the contribution of the Si substrate and deposited

Al2O3 are very intense compared to the GST signal. These background con-

tributions need to be subtracted from the measured diffracted signal. For this

purpose, the blank sample (Si + Al2O3 ) has been measured. The 2D image

obtained for the as-deposited GST film sample, after removing the background,

shows only faint traces, with no diffraction rings. This is a first confirmation that

the as-deposited GST film is amorphous. The 2D image of the 200C ex situ

annealed thin film sample, again after background subtraction, shows diffrac-

tion rings which can be indexed as a fcc structure with a (111) texture (Figure

3.11a).

Concerning the as-deposited clusters, the X-ray diffraction 2D image shows

no diffraction rings. For the 200C annealed clusters, supposing a cubic struc-

ture, the (111) and (222) diffraction rings are not observed but a weak signal

is visible for the (200) and (220) diffraction rings (Figure 3.11b), which are the

two most intense reflections expected in a powder diffraction pattern from the

bulk fcc phase. The intensity is constant over all the measured rings portions.

A diffuse scattering from the Si substrate is still present at the corners of image

Fig.3.11b, showing that the background subtraction is not perfect1.

1This could be related to the fact that the relative contributions of the Al2O3 signal and

Si signal are not the same in the blank sample and in presence of GST. This problem could

82


Figure 3.11: (a) X-ray 2D diffraction images for 200C ex situ annealed GST thin film. (b)

Same measurements for 200C ex situ annealed clusters. In both cases, the 2D image of the

blank sample (Si+Al2O3 ) has been subtracted.

The 2D images give information about the samples texture and traces of

Bragg peaks can be detected directly on them, but the evidence of the phase

transformation and the position for the diffraction lines are more clearly ob-

served by measuring the angular integrated intensity, in particular for clusters.

The intensity as a function of 2θ has been obtained by integrating over the en-

tire 2D images for each sample, excluding only a border of 100 pixels on each

side, thus losing information on texture in the film case. The background inte-

grated intensity has always been subtracted. The angular integrated intensity

for the film sample is shown in Figure 3.12. For the as-deposited film the clear

observation of two broad maxima at 2θ = 12.9 and 21.6 allows to conclude

that as-deposited clusters are amorphous. Their positions, corresponding to

Q = 4πsinθ/λ =2.03A−1 and 3.38 A−1, match the first two maxima of the GST

amorphous structure factor reported in literature [27] (see Chapter 2). For the

annealed film the peaks positions indexed in the fcc cubic structure indicate a

lattice parameter of 6.01 ± 0.01 A that closely matches the one reported for

bulk GST, 6.0117(5) A [25]. It should be noticed that due to the texture and

the finite angular range the relative integrated intensity for the peaks is not the

one expected for a powder pattern. Such intensity is reported in Table 3.2 as

calculated considering the lattice parameter for GST given in Ref. [25].

not be solved after the experience took place.

83


8 10 12 14 16 18 20 22 24

0

1x107

2x107

3x107

4x107

220

200

222

111

As-deposited

200°C

Inte

nsi

ty [

nu

mb

er

of

cou

nts

]

2q[°]

311

Figure 3.12: X-ray diffraction spectra at room temperature for as-deposited and 200C ex

situ annealed GST film, after background subtraction, with curves shifted for clarity. Ar-

rows indicate bulk GST fcc peak positions calculated assuming the lattice parameter of GST

a=6.0117 A reported in Ref.[25].

84


12 14 16 18 20 22

0.0

2.0x106

4.0x106

6.0x106

8.0x106

1.0x107

1.2x107

220200

As-deposited

200°C

Inte

nsi

ty [

nu

mb

er

of

cou

nts

]

2q[°]

Figure 3.13: X-ray diffraction spectra at room temperature for as-deposited and 200C ex

situ annealed GST clusters after background subtraction. Curves are shifted for clarity. Ar-

rows indicate bulk GST fcc peak positions calculated assuming the lattice parameter of GST

a=6.0117 A reported in Ref.[25].

85


hkl 2θ[] d[A] Irel

111 11.518 3.47086 4.25

200 13.307 3.00585 61.18

220 18.862 2.12546 100.00

311 22.155 1.81260 8.62

222 23.154 1.73543 57.35

Table 3.2: Peak positions and relative intensities for the fcc GST phase as expected from a

powder pattern. They have been estimated in a θ −2θ geometry at the actual experimental

wavelength considering the lattice parameter a=6.0117 A reported in Ref [25].

The angular integrated intensity obtained for ex situ clusters samples is re-

ported in Fig. 3.13. The as-deposited sample shows no clear trace of the amor-

phous broad maxima that can be identified for the as-deposited film sample,

as can be understood in view of the smaller quantity of GST in clusters sam-

ples. As expected, due to the smaller crystallite size, the peaks width for the

200C annealed cluster samples are larger than for the thin film. Besides, a clear

shift of the diffraction lines with respect to their position in the crystalline thin

film is observed.The fcc lattice parameter for the crystalline clusters calculated

from the (200) and (220) peaks position is 6.11 A ± 0.02.

In order to measure Tx more precisely the 2D X-ray diffraction spectra for

as-deposited thin films and clusters were recorded as a function of the tempe-

rature, with in situ annealing using a domed oven stage. More details on the

experimental setup and data treatment can be found in appendix A.2.2. These

measurements are challenging due to the temperature dependent spurious signal

from the oven dome that reaches more than 109 counts, two orders of magnitude

higher than the film signal and even three orders of magnitude higher than the

clusters signal. Nevertheless, the crystallization of the amorphous as-deposited

clusters could be observed. The temperature was increased by 10C steps after

which X-ray diffraction spectra were recorded for 26 min.

The change in the thin film (220) diffraction peaks from 150C to 180C is

plotted in Figure 3.14. Despite the deterioration of the signal over noise ratio

in presence of the dome covering the oven, it is clear that the peak intensity

86


18.0 18.2 18.4 18.6 18.8 19.0 19.2 19.4

0.0

5.0x106

1.0x107

1.5x107

2.0x107

Peak 220

Inte

nsi

ty [

nu

mb

er

of

cou

nts

]

2q[°]

140°C

150°C

160°C

170°C

180°C

200°C ex situ

220

Figure 3.14: (220) diffraction peak for in situ annealed GST thin film at different tempera-

tures. The dotted line is the peak, measured at room temperature, of the thin film annealed ex

situ at 200C (shifted for clarity). The arrow indicates the calculated bulk GST peak position.

87


12.5 13.0 13.5 14.0 14.5

0

1x106

2x106

3x106

4x106

5x106

17.5 18.0 18.5 19.0 19.5 20.0

(a)ex situPeak 200

Inte

nsi

ty [

nu

mb

er

of

cou

nts

]

2q[°]

150°C

170°C

190°C

210°C

230°C

(b)ex situ

Peak 220

2q[°]

220200

Figure 3.15: X-ray diffraction spectra for in situ annealed GST clusters at different temper-

atures. (a) (200) diffraction peak and (b) (220) diffraction peak. Curves are evenly shifted to

ease viewing. Dotted lines indicate the peak position, measured at room temperature, of the

200C ex situ annealed clusters.

measured at 180C is comparable with that measured at room temperature for

the sample annealed ex situ at 200C . At 150C no signal is recorded above

background level, while at 170C the crystalline peak is close to its final ampli-

tude. It can be observed that the (220) peak position at 180C is slightly below

that of the ex situ annealed thin film measured at room temperature, which can

be explained by the GST thermal dilatation [93, 94]. Considering a coefficient

of thermal expansion for Ge2Sb2Te5 of αT = 1.81 ·10−5K−1 [94] and the relation

d = d0(1 + αT ·∆T) for the thermal expansion of the interatomic distances, 2θ

is expected to vary of 0.05 between the room temperature and 200C .

The (200) and (220) diffraction peaks for in situ annealed clusters are shown

in Figure 3.15. Despite the high noise level and remaining contribution from the

oven dome at 2θ=19.5(Fig. 3.15b), the parallel rise in the amplitude for the

two peaks is visible. In order to make a more quantitative analysis the area of

the peaks has been integrated as a function of temperature2. Considering the

2Integration is restricted to the left part of the 220 peak in order to avoid the remaining

spurious signal of the dome.

88


Figure 3.16: Normalized integrated intensities for (220) and (200) diffraction peaks for GST

clusters and for GST film as a function of temperature. The normalized integrated intensities

have been obtained through the relation Inorm = I/Imax where I is the measured integrated

intensity at a given temperature and Imax is the maximum value of the integrated intensity

(which correspond to complete crystallization). The dotted lines indicate the crystallization

temperatures.

different values of the peak maxima for film and clusters samples, the obtained

integrated intensities have been normalized through the relation Inorm = I/Imax

where Inorm is the normalized integrated intensity and Imax is the maximum

value of the integrated intensity (which corresponds to complete crystallization).

All these intensity are measured at a given temperature. Inorm corresponds to

the crystalline fraction, thus allowing a direct and clearer comparison between

clusters and film. The crystallization temperature is defined as the temperature

corresponding to the midpoint of the rise step of Inorm. The results are shown

in Figure 3.16.

For the thin film the rise is almost parallel for the (220) and the (200) peaks.

No crystallization occurs at temperature lower than 140C . The value of Tx is

155C and the crystallization is completed at 170C . For the clusters, crystal-

lization starts at around 150C , Tx is close to 180C and the rise in amplitude

is more gradual, spanning from 150C up to 200C .

89

3.4 Discussion

3.4 Discussion

The most important result obtained from the ex situ cluster samples is the

confirmation that the nanoparticles are able to switch from the amorphous to

the crystalline fcc phase when annealed. So, even for particles with a diameter

as small as 5 nm, the phase change mechanism still takes place. This result is

positive for the future developments of Phase Change Memories.

A second key observation regarding the annealed crystalline clusters is that

the lattice parameter obtained from the measurements is larger than that of

the annealed fcc thin films, the latter being very close to the one expected for

the GST fcc cubic phase. For thin films, Scherrer analysis gives a lower limit

for the crystallite size [95] close to the layer thickness and a small inhomoge-

neous strain of 0.006. These films are close to complete relaxation. On the

other hand, the shift in the crystalline clusters peaks position indicates that the

lattice parameter for the fcc clusters (6.11A) is 1.7% larger than that of the

crystallized thin film (6.01A), which can be attributed to a large tensile strain

due to the interaction with the oxide matrix3. At the phase change transition,

the GST density decreases by 5% [25, 94]. In thin films, the stress resulting from

the volume change can be relaxed along the two space directions that are not

stuck to the surrounding layers, whereas clusters are bound to the surrounding

matrix in all three dimensions. Upon crystallization, the deformation must be

accommodated one way or another. This can be for instance via the creation

of voids [96, 97] or vacancies [98] or strain. In the present case, the observed

homogeneous tensile strain is very close in absolute value to the reduction in size

that would be expected for freestanding clusters during the amorphous to crys-

talline phase change. Indeed, the variation of the lattice parameter estimated for

clusters compared to the thin film ∆ aa

≈ 1.7% corresponds to a relative volume

variation ∆ VV

= (∆ aa)3 ≈ 4.9%. This suggests the scenario represented schemat-

ically in Fig.3.17. When an amorphous cluster switches to the crystalline phase

the strong interaction with the embedding alumina, which is a far more rigid

3In crystalline clusters, there are only two peaks that can be analysed. A Sherrer analysis

with no inhomogeneous strain leads to a grain size in agreement with the clusters diameter.

Therefore, inhomogeneous strain effects in clusters will not be considered.

90

3.4 Discussion

Figure 3.17: The Al2O3 matrix surrounding the cluster forces its volume to remain equal

to the one of the amorphous phase even after crystallization. The cluster in its amorphous

phase occupies a certain volume (a). When crystallization occurs, the cluster volume tends

to reduce of around 5% (b), but the embedding Al2O3 matrix exerts a tensile strain over the

cluster (c) thus forcing the cluster to keep the volume corresponding to the amorphous phase,

with the effect of increasing the lattice parameter of the crystalline cluster (d).

material than GST, forces it to keep the volume that it had in the amorphous

phase.

As already stated, there are no clear reported values of the crystallization

temperature for GST nanoclusters of such small diameter in literature. Never-

theless, it seems quite natural to compare our results with those reported in Ref.

[70] on GST thin films embedded in Al2O3 , which is the same cladding mate-

rial used in this study. The Tx obtained for the thin film in this work (155C )

is in agreement with the value reported in Ref.[70] for a GST film of similar

thickness. On the other hand, the Tx in the case of clusters is only 25C above

Tx of the 10 nm thick GST thin film, an effect much smaller than reported in

Ref.[70] where a thin film of 5 nm shows a Tx of more than 330C . However, the

different surface to volume ratio can have a deep influence on the crystallization

mechanism, making difficult a direct comparison between clusters and thin films

scaling properties. Moreover, a film is free to change its volume upon phase

transformation while clusters are confined in three dimensions and strained, as

shown above. The strain may possibly play a role in promoting or impeding

crystallization. However, no data about the deformation or strain of measured

films are available in Ref. [70], thus impeding to compare strain effects. In Ref.

[72] an effect of stress over crystallization has been addressed, but the variation

on lattice parameters due to strain has not been measured so again a comparison

is hazardous.

91

3.4 Discussion

As shown in section 3.1.1, it is very difficult to discuss effects on the crys-

tallization temperature since a wide number of factors plays a role. However,

considering that in the present case the GST clusters, the GST film and the

Al2O3 capping have been deposited in the same apparatus and measured in

identical conditions it is possible to try to interpret the observed 25C variation

of Tx between thin film and clusters. It could be related to many different factors

including composition effect, different surface to volume ratio, matrix influence,

stress or strain effects or an intrinsic size effect. A possible composition effect

could arise from the fact that, as compared with the films, the clusters are Te-

depleted. For instance the crystallization temperature in Ge-Sb-Te thin films

with 10%-20% excess Sb is 15C higher than for GST films [99]. The crystal-

lization temperature for Ge2Sb2Te4 is 175C [100].Another effect could be the

stress resulting from the phase change. From a thermodynamical point of view

the elastic energy stored in clusters will reduce the driving force for the phase

transition. The kinetics for the phase change will be slowed and, during a tem-

perature scan, the transition temperature will increase. In the case of GST this

driving force is 200 MJ/m3, [101]. An order of magnitude for the elastic energy

resulting from the strain is close to 50 MJ/m3, considering the experimental

value of the bulk modulus available in literature [102], so it can induce signifi-

cant effects. Finally, the increase of Tx could be an intrinsic size effect due to

the impact of surface energy as explained through Zacharias model [71], even

if the effect is much smaller than what reported in Ref.[70] for GST thin films

capped with Al2O3 .

As already mentioned at the end of section 3.3, the crystallization of clusters

is more gradual than the one of the 10 nm film. This can be due to a disper-

sion of the crystallization temperatures of clusters. If Tx increases for reduced

dimensions due to a size effect, smaller clusters will have a higher Tx so the

crystallization temperature dispersion is a consequence of the size distribution.

However, it is also possible to have a dispersion of Tx even for clusters of the

same size, due for example to oxidation of the clusters at the Al2O3 interface

which can change the composition and influence Tx . It is also worth noting

that, supposing an homogeneous nucleation and considering that the clusters

92


size is quite close to the critical nucleus diameter (about 2 nm), a dispersion of

Tx could occur from cluster to cluster.


In conclusion, nanometric GST clusters were deposited by a sputtering gas-

aggregation technique, with a narrow size distribution around 5.7 nm. The

obtained results demonstrate that this synthesis technique offers new possibility

for the study of well calibrated and isolated clusters of phase change materials,

opening the way to a systematic study of nanoparticles deposited by sputtering.

It offers the same reduced dimensions and size distribution as chemical synthe-

sis but with the possibility of depositing ternary compounds (this is still very

difficult with chemical techniques) and with a much smaller risk of including

contaminants. Moreover, this method is close to the physical techniques used

for PC film deposition in the fabrication of PCM, thus giving information that

can be more directly exported to device fabrication. The as-grown clusters dis-

persed in alumina are amorphous and transform into a fcc crystalline phase at a

well defined crystallization temperature of 180C . This is the first unambiguous

observation of this phase change in GST clusters in the sub-10 nm range. The

crystalline clusters show a lattice parameter larger than that of bulk cubic GST,

which indicates a tensile stress that can be attributed to the interaction with the

alumina matrix. The large increase in Tx observed in thin GST films subjected

to large interface stress [72] is not seen in clusters. These results indicate that

the scaling effect on the crystallization temperature in phase change material

can be small and the role of interfaces in term of stress and interfacial energy

effects must be further studied.

All the results reported in this section have been obtained from a single

experience at the synchrotron that was done over the first set of GST clusters

samples obtained with this deposition method. This experience allowed to learn

how to improve the data acquisition and subsequent analysis. For the next

experiences the quantity of matter in clusters samples should be increased in

order to obtain more intense signals, improve the quality of the data and reduce

93


the effort for the data treatment that was difficult and time consuming. The

blank sample needs to be deposited with the same amount of cladding material as

the PC samples in order to avoid errors in subtracting the substrate contribution.

In the near future the same deposition technique will be used to deposit

GeTe nanoparticles. Besides, based on the results presented in Chapter 4, the

effect of different capping layers as Ta, TiN, W or SiO2 will be also investigated.

Smaller clusters could also be deposited, but their size distribution would be

larger. A further step will be the electrical characterization of clusters in order

to determine their electrical properties such as threshold voltage, retention time,

reset current and cyclability. Doped clusters can also be deposited in order to

study the effect of doping at small sizes, and to evaluate the impact of the

variability of doping concentration in such small systems. To go further, local

order investigations and TEM measurements can be done in addition to XRD

measurements in order to completely characterize the structure of clusters.

94

Chapter 4

Interface effect on crystallization

of PC thin films

95

4.1 Introduction

4.1 Introduction

As already mentioned in section 3.1.1, the material used as a capping layer

has been shown to have an influence on the crystallization mechanism of a

phase change very thin film. Even for films as thick as 30 nm the same Ge-

Sb-Te compound tested under the same conditions crystallizes at a different

temperature if interfaced with different materials such as silicon dioxide SiO2 ,

silicon nitride Si3N4, tantalum oxide Ta2O5, zinc sulfide ZnS or ZnS − SiO2

[66], or, even when the value of Tx is unchanged, the crystalline phase and

growth rates of the same PC material can be different if capped with Si3N4 or

ZnS − SiO2 [77]. For film of GST and GeTe of larger thickness, the effect of

interfaces is assumed to be absent and the measured Tx is considered as the

bulk crystallization temperature.

In this chapter a study on the effect of different interfaces on the crystal-

lization temperature of a PC material thin film will be reported. The value

of Tx will be measured through reflectivity measurements (section A.1) and X-

Ray Diffraction measurements (XRD) (sections 4.3 and 4.4), and a possible

interpretation of the results will be proposed at the end of the chapter (section

4.5). The samples tested though reflectivity consist in thin films of GeTe and

GST of various thicknesses encapsulated in three different materials, TiN, Ta or

SiO2 . Those test materials have been chosen as capping layers because they are

frequently employed in the fabrication process of devices (TiN and Ta for the

electrodes and SiO2 as insulating material), so they can be eventually integrated

in a cell. The structure of the samples, shown in Fig. 4.1, is the same for both

GeTe and GST samples. The XRD experiments have been performed on GeTe

thin films 100 nm thick only. All the phase change materials and capping layers

have been deposited by sputtering as described in appendix B. The measured

composition of the GeTe samples is Ge52Te48. The experimental procedures and

setups are described in details in appendix A.2.

96

4.2 Reflectivity measurements

!"#

!$# !%#

!"#$%&$'()'*

+",#-.#/0

!"#$%&$'()'*

12#0)'*(")3

+",#4#/0

!"5-#4..#/0 +)#-.#/0

!"#$%&$'()'*

12#0)'*(")3

+)#4#/0

12#0)'*(")3

!"5-#6.#/0

Figure 4.1: Structure of PC material thin films samples used in order to investigate the

interface effect on crystallization. The PC material can be either GeTe or GST, of various

thicknesses, sandwiched between (a) SiO2 , (b) TiN or (c) Ta. All the samples have been

deposited by sputtering as described in B.


The reflectivity as a function of temperature of 100 nm thin films of GST and

GeTe sandwiched between TiN, Ta or SiO2 is shown in Figure 4.2. The heating

rate was 10C /min for all the samples. The amorphous to crystalline phase

transition corresponds to an increase in the value of the measured reflectivity.

It is evident from Fig. 4.2 that for both GST and GeTe the phase change

occurs at similar temperatures when the film is sandwiched between TiN or

SiO2 , while the transition occurs at a higher temperature when the PC material

is interfaced with Ta. This is a surprising result. Even if an effect of the

encapsulating material on Tx has been observed in literature for films of 30 nm

[66] it was very small, while in the present case we observe a difference of more

than 20C even for films 100 nm thick. The crystallization temperatures Tx for

the six samples of Fig. 4.2, calculated as the point of maximum derivative

of the reflectivity curve, are reported in Table 4.1. The difference between

the crystallization temperatures obtained with the TiN or SiO2 interfaces is

small both for GeTe and GST samples, and it is within the uncertainties on

the sample temperature in the reflectometer 1. The resulting values of Tx for

GST interfaced with SiO2 and TiN are close to the Tx reported in literature for

1As described in details in appendix A.1, the temperature in the reflectometer is measured

on the heating plate. If the thermal contact between the heating plate and the sample is not

perfect the actual temperature on the sample can differ from the measured one.

97


150 160 170 180 190 200 210 220 230

0.0

0.2

0.4

0.6

0.8

1.0

120 130 140 150 160 170 180 190 200

0.0

0.2

0.4

0.6

0.8

1.0

(a)

Cry

sta

llin

e f

ract

ion

GeTe 100 nm / SiO2

GeTe 100 nm / TiN

GeTe 100 nm / Ta

(b)

Temperature [°C]

GST 100 nm /SiO2

GST 100 nm / TiN

GST 100 nm / Ta

Figure 4.2: Crystalline fraction as a function of temperature obtained from reflectivity

measurements for (a) GeTe and (b) GST 100 nm thin films sandwiched between TiN, Ta or

SiO2 heated at 10C /min. For both GeTe and GST thin films the amorphous to crystalline

transition occurs at a higher temperature when the film is sandwiched between Ta.

98


GeTe 100 nm Tx [C ] GST 100 nm Tx [C ]

SiO2 188 148

TiN 193 152

Ta 213 165

Table 4.1: Crystallization temperatures Tx of GeTe and GST 100 nm thick films sandwiched

between TiN, Ta or SiO2 as obtained from the reflectivity measurements of Fig. 4.2 for a

heating rate of 10C /min.

30 nm thick films of GST interfaced with SiO2 and ZnS-SiO2 characterized by

transmittance measurements and resistance measurements [66, 67]. They are

lower than what reported for TiN interfaced GST measured by EXAFS and

ellipsometry in Ref.[72] and for GST interfaced with Al2O3 and measured by

X-Ray Diffraction [70], but in this last case the heating rate is six times higher

than in our experiments. The values of Tx obtained for GeTe films interfaced

with SiO2 and TiN are in agreement with what reported in Refs.[23] and [44] for

thin films measured under the same condition as in this work at a heating rate

of 20C /min and 10C /min, respectively. A Tx of 175C has been reported

in Ref.[103] for resistance measurements on a GeTe film 50 nm thick interfaced

with SiO2 heated at 1C /s and a Tx of 170C has been reported in Ref.[104]

for a GeTe film 75 nm thick interfaced with polymethyl methacrylate heated at

23C /min and measured by optical transmission measurements.

The same reflectivity measurement has been repeated on samples of GeTe

30 nm and 10 nm thick interfaced with TiN, Ta or SiO2 and the results are

shown in Fig.4.3. For the 30 nm thick samples the change in reflectivity can

be easily identified for each kind of interface, and the difference in Tx observed

for the 100 nm thick films between the Ta interfaced samples and the others is

still clearly evident. For the 10 nm thick GeTe films the phase transition for the

SiO2 interfaced sample is still evident, but the interpretation of the reflectivity

measurement becomes difficult for the TiN and Ta interfaced samples due to the

high contribution of the upper layer of TiN or Ta, that could not be subtracted.

However, a phase change can still be observed for the TiN interfaced sample.

On the other hand, no reflectivity change could be detected for the Ta interfaced

99


GeTe 30 nm Tx [C ] GeTe 10 nm Tx [C ]

SiO2 191 197

TiN 187 211

Ta 223 >230

Table 4.2: Crystallization temperatures Tx of GeTe films 30 nm and 10 nm thick sandwiched

between TiN, Ta or SiO2 as obtained from the reflectivity measurements of Fig. 4.3 for a

heating rate of 10C /min.

sample in the experimental temperature range up to 230C . It could be due to

the fact that Tx for that sample is above 230C , but no definitive conclusion

can be drawn in view of the difficulties of the reflectivity measurements. The

values of Tx for the 30 nm and 10 nm thick GeTe samples are shown in Table 4.2.

For the 30 nm thick films, the difference of Tx between the Ta interfaced sample

and the others is around 30-35C , while it was around 20-25C for 100 nm thick

GeTe films. By comparing Tables 4.1 and 4.2 it can be concluded that reducing

the film thickness has a weak effect on Tx for the SiO2 interfaced samples, while

Tx seems to increase with deacreasing thickness for the samples interfaced with

TiN and Ta. Indeed, the 100 nm and 30 nm TiN interfaced samples have the

same Tx , while Tx increases of around 20C for the 10 nm thick film. The Tx of

Ta interfaced GeTe samples increases of around 10C when the film thickness is

reduced to 30 nm and no phase transition can be seen up to 230C for the GeTe

film 10 nm thick.

The activation energy EA of the GeTe 100 thin film has been calculated for

the SiO2 and Ta interfaced samples by repeating the reflectivity measurement

for three other heating rates: 3C /min, 5C /min and 20C /min. The crystal-

lization temperatures Tx obtained for all different heating rates are reported in

Table 4.3. The values of EA obtained by fitting the set of points shown in Fig.

4.4 (as described in section 1.3.5) are of 3.83 eV for Ta interface and 2.58 eV for

SiO2 interface. The activation energy obtained for GeTe interfaced with SiO2 is

close to the values known from literature [23]. EA is significatively enhanced for

the Ta interfaced sample.

100


150 160 170 180 190 200 210 220 230

0.0

0.2

0.4

0.6

0.8

1.0

150 160 170 180 190 200 210 220 230

0.0

0.2

0.4

0.6

0.8

1.0

(a)

Cry

sta

llin

e f

ract

ion

GeTe 30 nm / SiO2

GeTe 30 nm / TiN

GeTe 30 nm / Ta

(b)

Temperature [°C]

GeTe 10 nm / SiO2

GeTe 10 nm / TiN

GeTe 10 nm / Ta

Figure 4.3: Crystalline fraction as a function of temperature from reflectivity measurements

for (a) GeTe 30 nm and (b) GeTe 10 nm thin films sandwiched between TiN, Ta or SiO2 ,

heated at 10C /min. For 30 nm thick GeTe films the amorphous to crystalline phase transition

occurs clearly at a higher temperature when the PC material is sandwiched between Ta. For

10 nm thick films of GeTe the measurement becomes difficult for samples interfaced with TiN

and Ta, while Tx can be still easily identified for the SiO2 interfaced sample.

101


23.6 23.8 24.0 24.2 24.4 24.6 24.8 25.0 25.2 25.4 25.6 25.8

-11.5

-11.0

-10.5

-10.0

-9.5

-9.0

EA=2.58 eV

SiO2 interface

Ta interface

-ln

(r/T

x²)

1/(kBT

x) [eV

-1]

EA=3.83 eV

Figure 4.4: Kissinger plot for GST 100 nm thin films sandwiched between Ta or SiO2 . The

absolute value of the line slope corresponds to the activation energy EA . The points in graph

have been calculated from the Tx value obtained for four different heating rates r as reported

in Table 4.3

102

4.3 X-Ray Diffraction measurements

Heating rate Tx for SiO2 interface [C ] Tx for Ta interface [C ]

3 C /min 179 207

5 C /min 184 210

10 C /min 188 213

20 C /min 191 217

Table 4.3: Crystallization temperature Tx obtained from reflectivity measurements for dif-

ferent heating rates for GeTe 100 nm thin films sandwiched between Ta or SiO2 .

1

2

Detector

SampleBeam

Beam

source

Figure 4.5: Schematic representation of the experimental geometry of the XRD experiment,

where θ is the incident beam angle and Ψ is the tilting angle of the sample. A detailed

description of the experimental setup is provided in appendix A.2.1


Three as deposited 100 nm thick films of GeTe have been characterized by

in situ annealing XRD in order to investigate their crystalline structure. The

samples structure is the same as the one used for the reflectivity measurements

samples (shown in Fig. 4.1). The XRD experiments have been performed at

λ = 1.540598A using a PANalytical instrument equipped with a punctual pixel

detector in θ-2θ configuration as described in A.2.1. The samples have been

annealed under N2 atmosphere from 100C to 300C by steps of 10C . At

each step the temperature was kept constant for 5 minutes before starting the

measurement that lasted 1 hour and 25 minutes. Note that in the reflectivity

measurement the temperature as a function of time is a ramp and the heating

rate is defined as the constant slope of the temperature ramp. For the XRD

experiment the temperature as a function of time is a staircase function. In order

to compare the XRD results and the reflectivity measurements results, it is useful

to define an equivalent heating rate approximated as equal to the step increase

divided by the time step (0.11C /min). Each measurement was performed in

103


25 30 35 40 45 50 55

0

100

200

300

400

500

600

700

800I(

2q)

[#

of

cou

nts

]

2q [°]

SiO2

Ta

TiN

Rhomboedric GeTe

20

2

02

10

06

11

3

01

5

11

0

10

4

10

1

00

3

01

2

Figure 4.6: Diffracted intensity as a function of 2θ for Ψ =0of GeTe 100 nm thin films

sandwiched between TiN, Ta or SiO2 measured at 100C after annealing at 300C with a

heating rate of about 0.11C /min. The vertical lines correspond to the calculated position

of Bragg peaks for rhombohedral GeTe (hexagonal indexation) [25]. No difference in the

diffraction spectra can be observed between the Ta and TiN interfaced samples, while no

peaks are visible for the sample sandwiched in SiO2 . The intense peak around 33 is due to

Ta.

the range 23 < 2θ < 55 for 4 different values of the tilting angle Ψ, shown in

Fig. 4.5, corresponding to 0, 20, 40 and 60. The same measurements have

been done also for a descending temperature ramp after crystallization, from

300C to 100C , under the same conditions for temperature steps of 10C .

The intensity as a function of 2θ for Ψ=0 for the three samples measured at

100C after the annealing at 300C is reported in Fig. 4.6. In the same figure

the peak positions of rhombohedral GeTe at the experimental wavelength are

reported, estimated by considering the lattice parameters a = 4.164 A and c

= 10.69 A with an hexagonal indexation [25]. It is immediately evident that

the SiO2 interfaced sample shows no trace of Bragg peaks, except a weak peak

around 51. Considering that the same sample has been observed to be fully

crystalline after annealing at 230C through reflectivity measurements, the ab-

sence of Bragg peaks is probably an effect of texture, as it will be confirmed

104


in the next paragraph. For the Ta and TiN interfaced samples it is possible to

clearly identify the 101 and 012 peaks, as well as a strongly asymmetric peak

around 43. The peak shape suggests the presence of two peaks that cannot be

distinguished due to the measurement resolution and correspond to the 104 and

110 peaks of the GeTe rhombohedral phase. The intense peak around 33 in the

Ta interfaced film is due to Ta.

In order to check the samples textures at complete crystallization, the 012

peaks of all the samples measured at 100C after annealing at 300C have been

reported in Fig.4.7 for each tilting angle Ψ. It is immediately clear from the figure

that the SiO2 interfaced sample is strongly textured, since the peak intensity

varies strongly with Ψ. In particular, the peak intensity is maximum for Ψ=40,

very weak for Ψ=20 and equal to zero for Ψ=0, indicating that the angle

between the preferred orientation of the diffracting planes and the sample surface

is close to 40. The Ta and TiN interfaced samples show only a weak texture.

The ratio ∆A/Amean, where ∆A is the difference between the maximum and

minimum value of the peaks area and Amean is the mean area over all the values

of Ψ, is around 2.62 for the SiO2 interfaced sample, 1 for the TiN one and 0.5446

for the Ta one, confirming the difference in texture.

Considering that the most intense peaks for the SiO2 interfaced sample are

observed for Ψ=40, the evolution of the 012 Bragg peak of GeTe as a function

of temperature for each sample has been measured for Ψ = 40 and the results

are reported in Fig.4.8. The Bragg peak appears at around 150C for the sam-

ple interface with TiN, 160C for the sample interfaced with SiO2 and 190C for

the sample interfaced with Ta, confirming the difference of crystallization tem-

perature observed in reflectivity measurements (section A.1). The peak area

for each sample has been normalized between 0 and 1 to obtain the crystalline

fraction as a function of temperature, as reported in Fig.4.9, allowing the direct

comparison with the reflectivity data of Fig.4.2. The values of Tx are shown in

Table 4.4.

In the case of in situ XRD experiment for all samples the crystallization takes

place at lower temperatures compared to the reflectivity measurements due to

the different heating rates used in the XRD and reflectivity experiments.

105


28.5 29.0 29.5 30.0 30.5 31.0

0

200

400

600

800

1000

1200

1400

1600

1800

2000

2200

28.5 29.0 29.5 30.0 30.5 31.0

0

100

200

300

400

500

600

700

800

900

1000

1100

1200

1300

28.5 29.0 29.5 30.0 30.5 31.0

0

100

200

300

400

500

600

700

800

900

1000

1100

Psi = 0°

Psi = 20°

Psi = 40°

Psi = 60°

GeTe 100 nm / SiO2 10 nm

2q [°]

2q [°]

Inte

nsi

ty [

nu

mb

er

of

cou

nts

]

GeTe 100 nm / TiN 5 nm

Psi = 0°

Psi = 20°

Psi = 40°

Psi = 60°

Inte

nsi

ty [

nu

mb

er

of

cou

nts

]

GeTe 100 nm / Ta 5 nm

Inte

nsi

ty [

nu

mb

er

of

cou

nts

]

2q [°]

Psi = 0°

Psi = 20°

Psi = 40°

Psi = 60°

Figure 4.7: 012 Bragg peak for 100 nm thick GeTe films interfaced with SiO2 , TiN and Ta

measured at 100C after annealing at 230C (heating rate of about 0.11C /min) for various

tilting angles Ψ of the samples. The Ta and TiN interfaced samples show a weak texture

while the sample sandwiched in SiO2 is strongly textured, with a maximum peak intensity for

Ψ=40 and no intensity for Ψ=0.

106


28.0 28.5 29.0 29.5 30.0 30.5 31.0

0

200

400

600

800

1000

1200

1400

1600

1800

2000

2200

28.0 28.5 29.0 29.5 30.0 30.5 31.0

0

100

200

300

400

500

600

700

800

900

1000

28.0 28.5 29.0 29.5 30.0 30.5 31.0

0

100

200

300

400

500

600

700

800

900

1000

Inte

nsit

y [

nu

mb

er

of

co

un

ts]


2q [°]

120

130

140

150

160

170

180

190

200

Inte

nsit

y [

nu

mb

er

of

co

un

ts]



2q [°]

120

130

140

150

160

170

180

190

200

160

170

180

190

200

210

220

230

Inte

nsit

y [

nu

mb

er

of

co

un

ts]

2q [°]

Figure 4.8: Evolution of the GeTe 012 Bragg peak as a function of temperature observed for

Ψ=40 (heating rate of about 0.11C /min) for the 100nm thick GeTe films interfaced with

SiO2 , TiN and Ta. For each sample the thickest line in the graph corresponds to the first

temperature at which the Bragg peak becomes visible.

107


120 130 140 150 160 170 180 190 200 210 220 230

0.0

0.2

0.4

0.6

0.8

1.0

Cry

sta

llin

e f

ract

ion

Temperature [°C]




Figure 4.9: Evolution of the crystalline fraction as a function of temperature (heating rate

of about 0.11C /min) for the 100nm thick GeTe films interfaced with SiO2 , TiN and Ta as

obtained from the 012 GeTe Bragg peak area measured by XRD through in situ annealing for

a tilting angle Ψ=40.

In order to investigate a possible evolution of the samples textures with

temperature the area of the 012 Bragg peak has also been calculated as a function

of temperature for all the tilting angles. The results are shown in Fig. 4.10.

The crystallization temperature of each sample, calculated as the temperatures

corresponding to the midpoints of the rise steps of the plotted curves, remains

constant for all the values of Ψ and correspond to the values already reported

in Table 4.4. It can be noticed that at the beginning of crystallization the

relative peak intensities for different values of Ψ are the same as at complete

GeTe 100 nm Tx [C ]

SiO2 154

TiN 153

Ta 175

Table 4.4: Crystallization temperatures Tx of 100 nm thick GeTe films sandwiched between

TiN, Ta or SiO2 , obtained as the temperatures corresponding to the midpoints of the rise

steps of the Bragg peaks areas as a function of temperature reported in Fig. 4.9 (heating rate

of 0.11C /min).

108


100 120 140 160 180 200 220 240 260 280 300

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

100 120 140 160 180 200 220 240 260 280 300

-1000

0

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000

100 120 140 160 180 200 220 240 260 280 300

-1000

0

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000


Pe

ak

Are

a

Temperature [°C]

Psi = 0°

Psi = 20°

Psi = 40°

Psi = 60°


Pe

ak

Are

a

Temperature [°C]

Psi = 0°

Psi = 20°

Psi = 40°

Psi = 60°


Pe

ak

Are

a

Temperature [°C]

Psi = 0°

Psi = 20°

Psi = 40°

Psi = 60°

Figure 4.10: 012 Bragg peak area as a function of temperature (heating rate of 0.11C /min)

and of the sample tilting angle Ψ for 100 nm GeTe films sandwiched between TiN, Ta or SiO2 .

The points on the descending temperature ramp are also shown, and no evolution occurs during

the cooling down process.

109

4.4 Synchrotron X-Ray Diffraction

Ψ=0 [nm] Ψ=20 [nm] Ψ=40 [nm] Ψ=60 [nm]

SiO2 77 68

TiN 63 67 67 65

Ta 56 60 62 58

Table 4.5: Final mean grain sizes measured at 100C after annealing for different values of

the tilting angle Ψ for 100nm thick GeTe films.

crystallization only for the SiO2 interfaced sample. For both the TiN and Ta

interfaced samples, the most intense peaks at the beginning of crystallization

correspond to Ψ=40 and Ψ=60, while the most intense peak is found for

Ψ=20 at complete crystallization. This suggests an evolution of texture with

temperature.

The evolution of the grain sizes as a function of temperature, calculated us-

ing Sherrer analysis on the 012 Bragg peaks of all the samples and neglecting

any inhomogeneous strain (see appendix A.2.1), is shown in Fig.4.11. The in-

strumental resolution is FWHMinstr=0.295and the error bar for the grain size

is about ± 5 nm. The grain size has been calculated for each point of the in-

creasing temperature ramp and at 100C after annealing. For Ψ=0 no peaks

are detectable for the SiO2 interfaced samples and for Ψ=20 the signal is so

weak that no reliable results can be obtained, thus for Ψ=0 and Ψ=20 the

grain size has been calculated only for the TiN an Ta interfaced samples. When

possible, the grain size after in situ annealing has been calculated and is reported

in Table 4.5. For both Ψ=40 and 60, the sample interfaced with SiO2 has the

highest grain size, while the Ta interfaced sample has the smallest grain size and

intermediate values are found for the TiN interfaced sample. In particular, the

effect is more marled for Ψ=40. For both Ψ=0 and Ψ=20 the grain size is

slightly larger for the TiN interfaced sample than for the Ta interfaced one.


A set of samples of 100 nm thick GeTe films interfaced with Ta, TiN or SiO2 has

been annealed at 400C for 15 minutes (equivalent heating rate around 26.7C /min)

110


100 120 140 160 180 200 220 240 260 280 30030

35

40

45

50

55

60

65

70

100 120 140 160 180 200 220 240 260 280 30030

35

40

45

50

55

60

65

70

100 120 140 160 180 200 220 240 260 280 300

30

40

50

60

70

80

100 120 140 160 180 200 220 240 260 280 300

30

40

50

60

70

TiN

Ta

Gra

in s

ize

[n

m]

ψ=0°

Temperature [°C]

Temperature [°C]

TiN

Ta ψ=20°

SiO2

TiN

Ta ψ=40°

Gra

in s

ize

[n

m]

Temperature [°C]

SiO2

TiN

Ta

Gra

in s

ize

[n

m]

Gra

in s

ize

[n

m]

ψ=60°

Temperature [°C]

Figure 4.11: Grain size calculated by Sherrer analysis on the 012 Bragg peak of GeTe 100 nm

thin films interfaced with SiO2 , TiN or Ta. The error bar is about ± 5 nm. The dashed lines

indicate the final dimensions of grains, measured at 100C after annealing, and are reported

in Table 4.5.

111

4.5 Discussion and conclusions

and has been characterized by X-ray Diffraction on the BM02 CRG-D2AM

beamline (ESRF Grenoble, France) with an incident photon energy of 17.5 keV

photon (λ =0.707 A) and using a 2D CCD camera detector. A detailed descrip-

tion of the experimental setup, which is similar to the one used in Chapter 3,

can be found in appendix A.2.2. The 3D image of the SiO2 interfaced sample

shown in Fig. 4.12(a) clearly shows that the intensity of the partial diffracted

rings corresponding to GeTe 101, 012, 104 and 110 Bragg peaks is zero around

the center of the image and very high at the image borders, indicating a strong

texture. In Fig.4.12(b) the TiN interfaced sample is characterized by a light tex-

ture while the Ta interfaced sample in Fig.4.12(c) exhibit completely isotropic

rings. This result is in agreement with those obtained from the in situ XRD ex-

periments described in section 4.3 for the SiO2 interfaced sample. In particular,

the absence of signal measured at Ψ=0 agrees very well with what can be seen

in Fig. 4.12(a). However, the weak texture observed in section 4.3 for the Ta

sample is absent, while the texture of the TiN interfaced sample is stronger in

Fig.4.12(b) compared to what observed in section 4.3. This observation suggests

that the final texture is influenced by the anneling conditions. The samples were

identical in both experiments but at the synchrotron they have been annealed

at 400C for 15 minutes, while the samples of section 4.3 have been heated up

to 300C at a heating rate of 0.11C /min, staying at high temperature for a

long acquisition time.


An effect on the crystallization temperature of interfacing GeTe and GST films

with a Ta capping layer has been observed and confirmed both through XRD

and reflectivity measurements. First, the increment of Tx for a sample interfaced

with Ta compared to SiO2 or TiN interfaced ones exists even for films 100 nm

thick, indicating that it is not correlated to a size effect. Moreover, the activation

energy EA is higher for Ta interfaced samples than for the SiO2 interfaced ones.

The grain size measured for Ψ=40 is higher for the SiO2 interfaced sample

than for the TiN and Ta interfaced samples, and the SiO2 interfaced sample

112


Figure 4.12: 3D images of the diffracted rings obtained for GeTe 100 nm thin films annealed

at 400C for 15 minutes and interfaced with (a) SiO2 (b) TiN (c) Ta. The SiO2 interfaced

film is strongly textured while only a faint texture is visible for the TiN interfaced film and

the GeTe rings are isotropic for the Ta interfaced film. The strongly textured rings that can

be seen in (c) correspond to Ta.

113


SiO2 TiN Ta

GeTe 100 nm Tx [C ] 188 193 213

EA [eV] 2.58 not measured 3.83

Thickness dependence of Tx weak weak strong

Texture strong weak weak

Grain size for Ψ=40 [nm] 77 67 62

Table 4.6: Summary of the different characteristics observed for GeTe thin films encapsulated

in SiO2 , TiN or Ta. The SiO2 and Ta interfaced samples present the most relevant differences

in their properties, while the TiN interfaced sample can be considered as an intermediate

situation.

is strongly textured while it is not the case for the other samples. Finally, the

crystallization temperature remains almost constant for the SiO2 interface as the

PC films thickness is reduced from 100 nm to 10 nm, while it increases for the

the other samples, the effect being more marked on the Ta interfaced sample. All

these different characteristics are summarized in Table 4.6, where it is evident

that the samples interfaced with SiO2 and Ta exhibit relevant differences in their

properties while the TiN interfaced sample can be considered as an intermediate

situation. For this reason, the discussion will be focused mostly on the differences

between SiO2 and Ta samples.

In order to identify the preferred orientation in the SiO2 interfaced sample,

the diffracted intensity as a function of 2θ (already reported in Fig.4.6 for Ψ=0)

is shown in Fig.4.13 for each value of Ψ. The curves are vertically shifted for

clarity. It can be immediately noticed that while the 012 Bragg peak reaches

its maximum intensity for Ψ of about 40(see Fig.4.7), the 101, 202, 104 and

110 peaks exhibit the highest intensity for Ψ=20. It can be observed that the

angle between the (012) plane and the (010) plane is around 38, while the angle

between the (101) and (100) planes is around 21. This suggest that the grains

grow with a preferred orientation with the (100) or (010) plane parallel to the

sample surface.

It is worth noting that different values of EA and Tx could be explained by

an effect of doping due to the diffusion of the encapsulating material into the PC

114


24 28 32 36 40 44 48 52

0

500

1000

1500

2000

2500

Y =60°

Y =40°

Y =20°

Y =0°

20

2

02

10

06

11

3

01

5

11

0

10

4

10

1

00

3

01

2

Inte

nsi

ty [

nu

mb

er

of

cou

nts

]

2q [°]

Figure 4.13: Diffracted intensity as a function of 2θ for the GeTe 100 nm thin film interfaced

with SiO2 , measured at 100C after annealing at 300C (heating rate around 0.11C /min)

for various tilting angles Ψ. The intensity of the 012 Bragg peak is maximum at around

Ψ=40, as already reported in Fig.4.7, while the 101, 202, 104 and 110 peaks exhibit the

highest intensity for Ψ=20.

115


0 200 400 600 800 1000 1200

0

500

1000

1500

2000

2500

3000

3500

4000

0 200 400 600 800 1000 1200

0

1000

2000

3000

4000

GeTe 30 nm

Co

un

ts J

(1

/s)

Sputtering time A (s)

Ge+

Sb+

Te+

Ta+

GST 30 nm

Ge+

Sb+

Te+

Ta+

Co

un

ts J

(1

/s)

Sputtering time A (s)

Figure 4.14: Secondary Ion Mass Spectrometry (SIMS) measurements performed on GeTe

and GST 30 nm thin films sandwiched with Ta. By definition, the interface for each element

can be placed in the point at which half of the signal intensity is lost, corresponding to the

vertical lines in the figure. From those measurements the diffusion of Ta inside the GeTe and

GST layers is extremely low.

layer, as described in Chapter 2 in the case of N and C doping elements. In order

to exclude this effect, Secondary Ion Mass Spectrometry (SIMS) measurements

have been performed on GeTe and GST 30 nm thick films sandwiched in Ta.

The results, reported in Fig. 4.14, show a very weak diffusion of Ta inside GeTe

or GST films, thus establishing that the samples are not Ta-doped.

It is possible to find an explanation for the observed phenomena by making

the hypothesis that the SiO2 /GeTe interface is energetically more favorable for

a particular orientation than the Ta/GeTe interface. If this hypotesis is true,

it leads to some consequences that will be analyzed in the following. First, the

different grain dimensions obtained at Ψ=40for SiO2 and Ta interfaced samples

can be explained as a result of an abnormal grain growth along a preferred

direction [105]. If the interfacial energy between GeTe and SiO2 is minimized

for grain textured with the (012) plane tilted by around 40with respect to the

sample surface, then the the growth of grains with this orientation is preferred.

As a result, the grain sizes are larger for those grains and the sample is textured,

corresponding to what observed experimentally.

A second consequence of the hypothesis of an energetically favorable SiO2 interface

can be a different crystallization mechanism for the two samples. The crystal-

lization can begin as heterogeneous nucleation at the interfaces for the sample

116


sandwiched in SiO2 and then the growth of those nuclei is dominant in the

crystallization process, being more rapid than the homogeneous nucleation that

takes place in the bulk of the PC material. This situation in illustrated in Fig.

4.15 a. On the other hand, as represented in Fig. 4.15 b, for the sample inter-

faced with Ta the heterogeneous nucleation at the interfaces could be somehow

suppressed or slowed down, so that the crystallization process is now dominated

by the homogeneous nucleation which is known to be normally slower than the

heterogeneous nucleation (see section 1.3.3).

The difference in the crystallization mechanism between samples interfaced

with Ta and SiO2 is supported by the evolution of the crystallization tempera-

ture Tx with thickness. Considering the model of Zacharias reported in section

3.1.1, homogeneous nucleation can induce large variation of Tx with shrinking

thickness while it is not necessarily the case for heterogeneous nucleation. This

is coherent with what observed for Ta and SiO2 interfaced samples, as reported

in Table 4.6.

The situation of the sample interfaced with TiN is somehow in the middle

between the Ta and SiO2 interface cases. The value of Tx corresponds to the

one measured for the SiO2 interfaced sample, while the weak texture and the

grain size indicate no preferred orientation for grains growth. This suggests that

the crystallization begins with heterogeneous nucleation as in the SiO2 interfaced

sample case, but with no preferred grain orientation and no consequent abnormal

growth, leading to a weak texture of GeTe and a grain size comparable to the

one measured for the Ta interfaced sample.

Even if the assumption of an energetically favorable interface for SiO2 can

well explain the obtained results, some questions are still open. If the situation

depicted in Fig.4.15 is true, the SiO2 interface should promote heterogeneous nu-

cleation more than the Ta interface, which should inhibit the formation of nuclei

instead. This is unexpected because it looks unlikely that an amorphous could

trigger nucleation more easily than a metal. Indeed, in Ref.[12] the nucleation of

various Ge-Sb-Te compounds have been studied and no heterogeneous nucleation

at the PC material/substrate interface was observed, even for a SiO2 interface,

and the only nucleation observed is the one at the interface with the native oxide

117


Figure 4.15: Possible models of crystallization for (a) SiO2 and (b) Ta interfaced GeTe

thin films. Different colors correspond to different orientation of the grains. In the case of

SiO2 interface the crystallization begins with heterogeneous nucleation at the energetically

favorable interface and the nuclei grow with a preferred orientation before the homogeneous

nucleation starts. The final result are bigger grains with a preferred orientation. In the

case of Ta interfaced PC thin film, the heterogeneous nucleation at the interfaces is somehow

suppressed, thus the crystallization is driven by the homogeneous nucleation that starts later

respect the heterogeneous one, leading to a weak texture.

118


of the PC material. The homogeneous nucleation was not observed either. This

can be explained by considering that the native oxide for a Ge-Sb-Te material

is usually GeO or GeO2 so that the Ge-Sb-Te material is Ge-depleted at that

interface and this lowers Tx [6]. In the case of samples interfaced with a capping

material on both sides, as the films studied here, no native oxide should exist at

the interfaces so this preferential nucleation site is absent. However, considering

that the film of SiO2 is deposited by PVD, it can present an excess of Si or

O. In case of an O-rich SiO2 an oxidation at the GeTe/SiO2 interface cannot be

excluded, while in the case of a Si-rich SiO2 some nanocrystals of Si can possibly

form inside the oxide and act as nucleation sites [106].

Even if the effect of different interfaces on the Tx of a PC material is ev-

ident, the reasons behind this phenomena are still unclear. Further studies

are required, focused on confirming the hypothesis of an energetically favorable

grains orientation at the GeTe/SiO2 interface and the following consequences

on the nucleation and growth.Besides, studies by XPS experiments on the local

structure at the interfaces are required. TEM measurements could be useful in

order to investigate nucleation mechanism for different interfaces.

119

Conclusion

The present work gives a contribution to understand the properties of some

PC materials used in PCM devices and the effects of scaling and interface on

the amorphous to crystalline phase transformation.

The first part of the thesis has been dedicated to investigate the local struc-

ture of C and N doped amorphous GeTe. The aim was to understand the local

origin of better electrical properties of doped GeTe devices compared to un-

doped GeTe ones. The impact of doping was observed experimentally through

the appearance of a new peak in the pair distribution function of doped GeTe, in-

dicating the formation of a bond at a new distance that is absent in the undoped

amorphous material. The formation of new environments involving carbon and

nitrogen has been confirmed through ab-initio simulations. In the case of car-

bon doping, strong changes are induced at the second neighbor level through

tetrahedral and triangular units centered on carbon. The new peak observed

experimentally corresponds to new Ge-Ge distances in these units, while for N-

doping it correspond to the new Ge-Ge distance in tethahedral and pyramidal

units centered on nitrogen. It should be remarked that for C doped GeTe mea-

sured and calculated pair distribution functions are in good agreements, whereas

the agreement is not as good for N doped GeTe. One possible explanation, to

be confirmed, is that an important proportion of nitrogen form N2 molecules

in the film. Further steps require the study of the crystalline phase of doped

materials, in order to understand the role played by doping elements during

121

Conclusions

and after crystallization. This is extremely important from the point of view of

optimizing materials for a device.

The subject of the second part of this work has been the impact of con-

finement on GST crystallization. A fundamental requirement for further de-

velopment of PCM is their ability to be scaled without deterioration of their

properties. The results known from literature for thin films, reported in Chap-

ter 3, were not encouraging. In this thesis, nano-sized clusters of GST embedded

in Al2O3 have been made using a sputtering gas phase condensation source and

their crystallization has been studied through X-ray diffraction. The amorphous

to cubic crystalline phase transition has been unambiguously observed for clus-

ters. The crystalline clusters experience a tensile strain that can be ascribed to

the effect of the surrounding rigid Al2O3 matrix. The crystallization tempera-

ture of clusters is only slightly higher than that of a 10 nm thin film of GST

deposited under the same conditions. It is worth underlining that this result

is positive for PCM because it shows that the scaling effect on the crystalliza-

tion temperature in a phase change material can be small. Further studies are

required in order to clarify the origin of this difference in crystallization tempe-

rature, since many different effects can be involved. A composition effect (rising

from the fact that clusters are slightly Te-depleted compared to the thin film),

different surface to volume ratio, matrix influence, stress effects or an intrinsic

size effect could all play a role. The obtained results open new possibilities for

the study of nano-sized clusters of phase change materials. These particles have

been deposited by a method which is close to the one used for PCM thin films

deposition, thus giving information that can be easily exported to device fabri-

cation. It would be interesting to deposit smaller GST clusters, but with this

method their size distribution will be larger. Furthermore, it would be useful

to study if nanoclusters of other PC material, such as GeTe, show the same

evolution of Tx with scaling. Finally, a study of the variability of phase change

properties with size, i.e. identification of intrinsic vs extrinsic effects, should be

done. In addition, the electrical properties of clusters can be measured.

The third and last part of the thesis has been dedicated to the investiga-

tion of the interface material effect on the crystallization temperature of GeTe

122

Conclusions

and GST. It has been observed through reflectivity measurements that both

GeTe and GST interfaced with Ta show a crystallization temperature around

20C higher than the one of GeTe and GST interfaced with TiN or SiO2 . Even

if some studies in literature had evidenced the influence of interfaces over the

crystallization temperature of Ge-Sb-Te materials, such a remarkable difference

in Tx , due only to an interface effect, was never reported before. X-Ray diffrac-

tion results on GeTe showed that the SiO2 interfaced samples are characterized

by a strong texture while a weak texture is observed for Ta and TiN interfaced

samples. The results obtained for Tx and the samples texture can be explained

by supposing different nucleation and growth mechanisms for the different sam-

ples. Nucleation would begin at the SiO2 /GeTe interface and grains grow along

a preferred direction through the film thickness, originating a textured crys-

talline film. For Ta interfaced samples the nucleation at the Ta/GeTe interface

would be somehow suppressed and the nucleation would be only homogeneous

with no preferred orientation and no texture. The case of the TiN interface is

somehow intermediate. However, it is unclear how an amorphous interface can

promote nucleation of crystal phase more easily than a metallic interface. Ac-

tually, a simulation in COMSOL Multiphysics is already in progress in order to

quantify the crystallization parameters that are influenced by different interface

materials.

From a more general point of view, studies should move from a global to

a local approach. It will be interesting to check the nucleation mechanism for

different interfaces not only through simulation, but also through transmission

electron microscopy measurements. This technique will be useful to observe

the phase change in nanoclusters of Chapter 3. Similarly, X-ray photoelectron

spectroscopy can be used to study the chemical bonding near the interfaces and

to investigate the chemical bonding of C and N doped GST and GeTe samples,

both in the amorphous and crystalline phase.

Finally, let us recall that the goal of these studies is the understanding of

scaling and confining effect on advanced phase change materials used in PCM. In

this thesis, the effects of doping, scaling and interfaces have been addressed sep-

arately. The obtained results and the success of the cluster deposition method

123

Conclusions

open new perspectives. For instance, considering the remarkable interface effect

evidenced in thin films, it would be interesting to deposit and characterize dif-

ferent PC materials clusters embedded in different matrices as Ta, TiN, W or

SiO2 . Indeed, such nanoparticles offer a high surface to volume ratio that can

enhance the interface effect. Another perspective is to deposit N or C doped

clusters. This could allow to study the impact of doping on such small systems,

as well as the effect of doping concentration variability.

124

Appendix A

Experimental Techniques

125

A.1 Reflectivity measurements Conclusions

Figure A.1: Evidence of the different optical properties of the PC material Ge2Sb2Te5 in

the amorphous and crystalline phases. The reflectivity of GST is reported as a function of

temperature starting from an initially amorphous sample. The amorphous phase is character-

ized by a low reflectivity value compared to the one of the crystalline phase. On the graph it

is easy to identify the crystallization temperature at which the phase transformation occurs.

A.1 Reflectivity measurements

Reflectivity measurements consist in monitoring the reflectivity value of a PCM

sample as a function of temperature. The amorphous phase is characterized by

a lower reflectivity than the crystalline phase, so when the phase transforma-

tion occurs the measured reflectivity value increases. An example of a measured

curve has been shown in the Introduction of Chapter 1 (Figure 1.2). The crys-

tallization temperature can be defined as the point of maximum derivative of

the measured reflectivity curve (other definitions of Tx can be considered, as

discussed in section 3.1.1). The measured reflectivity can be normalized in or-

der to obtain the crystalline fraction.

A schematic description of the reflectometer used for the experiments per-

formed in this thesis is reported in Figure A.2. An amorphous sample is placed

in a vacuum chamber on a heating plate and the sample reflectivity is mea-

sured through a red laser beam (λ=670 nm). The plate temperature can reach

126

Conclusions A.2 X-Ray Diffraction

Figure A.2: Schematic representation of the reflectometer used for reflectivity measurements.

The laser beam is directed onto a birefringent filter and divided in two beams, and one of them

is directed to the sample. The direct and reflected beams are collected by a photodetector

and processed to obtain the measured signal.

a maximum value of 400C with a heating rate that can vary between 2C /min

and 20C /min. The temperature is continuously measured by a thermocouple

and controlled by a temperature feedback. The samples can be measured under

vacuum or in an argon atmosphere, in order to limit the oxidation at high tem-

peratures. It is important to remark that the crystallization can be observed

only for the part of sample in which the laser beam penetrates (around 30 nm).

The capping layer that is deposited over the phase change material should then

be sufficiently thin to let the beam reach the underlying layer. Moreover, the

effect of beam reflection by the capping layer can affect the measured reflectivity

and can even hide the PCM signal if the capping material is not transparent

enough to the laser beam.

A.2 X-Ray Diffraction

From the analysis of X-ray diffraction it is possible to deduce information on

the sample structure and microstructure.

When an incident wave interacts with the electrons of an atom, if no energy loss

occurs, the result is a new spherical wave with the same energy of the incident

127

A.2 X-Ray Diffraction Conclusions

1

2

Detector

SampleBeam

Beam

source

Figure A.3: Schematic representation of the geometry used for XRD analysis in laboratory.

The experiment is performed in a θ-2θ configuration (θ1 = θ2) and the tilting angle Ψ allows

to measure the sample texture.

wave (elastic scattering). When two or more atoms are involved, the resulting

spherical scattered waves interact by constructive or destructive interference

depending on their phases.

In this thesis, the XRD experiments have been performed by means of two

different X-Ray sources. The first one is a conventional source, an X-Ray tube

used in laboratory measurements. The second one is the synchrotron radiation

that is required for measuring very small systems (as the nanoclusters described

in Chapter 3) due to its high brilliance.

A.2.1 Conventional X-Ray Diffraction laboratory exper-

iment

In laboratory, the XRD experiments were performed in the θ-2θ configuration

with the geometry schematically reported in Fig.A.3. The variation of the tilting

angle Ψ allows to measure the sample texture. The instrument consists in a

XPERT PRO MRD diffractometer equipped with a Cu anode (λ=1.5406A) and

a PANalytical pixel point detector. A furnace (Anton Paar) is used for XRD

with in-situ annealing in the temperature range 100−300C . In order to get

the peak width and position the diffracted peaks are fitted by a Pseudo-Voigt

function. Supposing that there is no inhomogeneous strain effect, the grain size

can be calculated from the peak width by using the Scherrer’s formula

Dhkl =0.9 · λ

(Γmeas − Γinstr) cosθ(A.1)

128


where Dhkl is the grain size calculated for the direction perpendicular to the

〈hkl〉 plane, λ is the experimental wavelength, Γmeas is the measured Full Width

Half Maximum, Γinstr is the instrumental Full Width Half Maximum and 2θ is

the peak position. For the instrument used Γinstr = 0.295.

A.2.2 Large-scale facilities experiments

In this thesis, experiments have been performed at two different Synchrotron

beamlines, the CRISTAL beamline (SOLEIL, Saclay) for the experiments de-

scribed in Chapter 2 and the BM02 CRG-D2AM beamline (ESRF, Grenoble)

for the experiments reported in Chapters 3 and 4 respectively. In both cases

the intensity as a function of 2θ is obtained by integration over a bidimensional

image but the experimental methods differ. On beamline CRISTAL X-ray scat-

tering has been measured up to large scattering angles in order to determine

the pair distribution function, while at ESRF selected Bragg peaks have been

studied.

SOLEIL data analysis

The experiment at the synchrotron SOLEIL have been performed in transmission

geometry over the powder samples described in Chapter 2. The experimental

setup is shown in Fig.A.4. The incident beam has an energy of of E=45.4793

keV (λ = 0.4441A) and the transmitted scattered beam is collected by an image

plate detector MAR350 (3450x3450 100µm pixels). An example of image is re-

ported in Fig.A.5 for an amorphous GeTe sample. The distance D between the

sample and the image plate was equal to ≈ 21cm and could not be reduced (see

Fig.A.4). Considering that obtaining a good quality structure requires measure-

ments in a large range of Q, and considering that the distance D could not be

reduced further, one chose a configuration where the center of the image plate

does not coincide with the incident beam as can be seen in Fig.A.5. This allows

to obtain a wider range of Q (up to 20.1 A−1) by integrating over a vertical

sector. One drawback of this set-up is a loss of intensity since only a fraction of

the diffraction ring is selected for integration.

129


Figure A.4: Picture and schematic representation of the experimental setup used at the

synchrotron SOLEIL. The scattered transmitted beam is collected by an image plate detector

placed at a distance D≈ 21cm from the sample, which is the minimum allowed distance in

this configuration. Thus, in order to obtain a high value of Q, the center of the image does

not correspond to the center of the detector.

130


The counting time for one image was 300 s. Each image has been corrected by

subtracting the dark image acquired without beam in the same conditions. All

the images have been treated by using the freeware software Fit2D [107]. It is

necessary to first correct the image for the geometrical errors due to the tilting

of the image plate with respect to the plane perpendicular to the incident beam.

The tilt parameters as well the distance between the sample and the detector

can be obtained by analysing the image of a LaB6 powder sample measured in

the same conditions. Each image can then be integrated. The integration sector

is shown in Fig.A.5. The integration process takes into account polarization

corrections. The value of polarization was 0.96.

The resulting intensity as a function of 2θ must finally be multiplied by 1−e(−dµIPcos2θ )

1−e−dµIP

in order to take into account the incomplete absorption of photons in the active

layer of the image plate of thickness d and absorption coefficient µIP .

In order to extract the scattering from the sample from the intensity measured

on the filled capillary, it is necessary to substract the scattering by the empty

capillary, taking into account the sample transmission, and to remove the scat-

tering by air. For each case (filled capillary, empty capillary or air) 4 images

have been acquired for 300 s and the obtained integrated intensities summed up.

The integrated intensity scattered by the sample I(2θ) is obtained by using the

following relation

I(2θ) = [(Iraw − Iair)− t · (Iemptycap − Iair)] ·1− e(−

µIPcos2θ )

1− e−µIP(A.2)

where Iraw is the raw intensity measured on the filled capillary , Iair is the air

signal, Iemptycap is the intensity of the empty capillary and t is the measured

transmission coefficient. The obtained quantity has then been processed by

using the PDFgetX2 software [108] which removes the Compton and fluorescence

signals, converts 2θ in Q and by adequate normalization gives the structure factor

S(Q). It should be emphasized that all the data corrections and normalization

must be performed carefully since any inadequacy directly affects the asymptotic

behaviour of S(Q). If S(Q) is not tending to 1 at large Q the pair distribution

function obtained by Fourier transform of Q(S(Q)-1) (see Eq. 2.18 in Chapter

2) is dramatically affected.

131


Figure A.5: Image acquired on the capillary containing amorphous GeTe. Only the pixels

contained in the vertical triangular sector shown on the figure have been selected for integra-

tion. The pixels that appear as a white spot at the center of the image plate are damaged

and must be avoided.

132


Figure A.6: Schematic representation of the experimental setup used at the synchrotron

ESRF. The diffracted beam is collected by a CCD camera at a distance D from the sample

of around 20 cm. The camera and the sample are tilted of an angle θ2 and θ1 with respect

to the incident beam, respectively. If θ1 = θ2, the configuration is in a strict θ −2θ geometry

only at the center of the camera.

ESRF data analysis

The diffraction experiments at the ESRF synchrotron (beamline BM02 CRG-

D2AM) have been performed in a θ-2θ configuration or in pseudo-grazing inci-

dence, over thin film samples and clusters samples described in Chapters 4 and

3.

The experimental setup is shown in Fig.A.6. The incident beam is directed

onto the sample, and the diffracted beam is collected by a CCD camera placed at

a distance D from the optical center. The CCD camera is a screen of 1340x1300

squared pixels of 50 µm. The angle between the camera and the incident beam

(θ2) has been chosen to be 17.5for all the performed experiments, while the an-

gle between the incident beam and the sample surface (θ1) was 8.75for the θ-2θ

configuration and 4for the pseudo-grazing configuration used in Chapter 3. The

incident angle of 4cannot be strictly considered as a grazing incidence, but for

smaller angles, the beam spot over the sample would have been larger than the

133


sample size (between 25nm2 and 400nm2). The diffracted beams are collected in

an angular range ±α in the horizontal direction and ±β in the vertical direction.

Strictly speaking, for θ1=8.75the geometry is in a real θ-2θ configuration only

at the center of the image. For all other points, the measured intensity collected

by the CCD camera is not due to diffraction of planes parallel to the sample

surface. In particular, at point B of Fig.A.6 is collected the beam diffracted

by a plane rotated by β/2 around an axis perpendicular to the figure plane.

Considering that the CCD camera is almost squared, α ≈ β and their values

depend on the distance D that determines the angular aperture measured by the

camera. For example, for the experiment on cluster described in Chapter 3 the

signal is measured in a 2θ range from 8 to 26, meaning that α ≈ β ≈ 9. The

exact value of the distance D is calculated for each experiment by measuring the

position of the direct beam on the camera without sample for θ2 values varying

between -6and 6.

For the experiments described in Chapter 3 on clusters and thin films, the in-

tensity as a function of 2θ has been obtained from each bidimensional image by

integrating over the rings (excluding only a border of 100 pixels on each side of

the image). A dedicated program, developed on the BM02 beamline, has been

used for integration. This program takes into account geometrical corrections,

in particular the fact that the camera plane is not vertical so that the diffrac-

tion rings are not circles. The resulting angular integrated intensities includes

the contribution of all the planes contributing to the measured partial rings on

the image, so any information on the texture of samples between this angular

aperture is lost in the final result. This means that the final integrated peak po-

sitions and widths are the sum of the peaks over the various Ψ, so care must be

taken in analyzing their width [95]. However, for non textured samples, as the

clusters characterized in Chapter 3, the rings are isotropic and no information

is lost with integration.

The samples described in Chapter 3 (GST thin film and nanoclusters) have

been studied by XRD as deposited, after ex-situ annealing and also during in

situ annealing experiment. In all cases, in order to get the signal of the studied

material it is necessary to subtract the contribution of the Si substrate and of

134


the Al2O3 encapsulating material from the measured signal. For this purpose,

the signal of a blank sample constituted of a 20 nm thick Al2O3 film deposited

on a Si substrate has been measured. However, the intensity of the substrate

and Al2O3 signal in sample measurement is not exactly the same as the one

measured for the blank sample, meaning that the intensity diffracted by GST,

IGST , should be obtained through the relation

IGST = Imeasured − αISi+Al2O3

(A.3)

where Imeasured is the intensity measured for the sample thus including the sub-

strate and Al2O3 contributions, ISi+Al2O3

is the intensity measured for the

blank sample and α is a coefficient that needs to be adjusted for each sample.

The fact that α is not equal to 1 can be due to many reasons, including a dif-

ference in the quantity of Al2O3 matter between the samples. It is important to

determine α accurately since ISi+Al2O3

is much larger than IGST . However, a

perfect subtraction of the substrate signal could not be achieved. This is prob-

ably due to the fact that the relative contributions of the Al2O3 signal and Si

signal are not the same in the blank sample and in presence of GST.

For the in situ experiments of Chapter 3, the furnace used is the same Anton

Paar furnace as for laboratory experiments (section A.2.1), equipped with a

PEEK dome. This dome gives an extremely intense undesired diffracted signal,

which is two orders of magnitude higher than the GST film signal and even three

orders of magnitude higher than the GST clusters signal.

135

Appendix B

Deposition method

137

Deposition method Conclusions

All the thin film studied in this thesis have been deposited by Physical Vapor

Deposition (PVD) through sputtering. In this method, a thin film is deposited

on the surface of a substrate by condensation of a vapor. The vapor is obtained

by bombarding a solid target (cathode) with a plasma and the ejected material is

directed onto the substrate (anode). In order to confine plasma on the surface of

the target, sputtering sources are equipped with magnetrons that utilize strong

electric and magnetic fields to increase the number of ionizing collisions near the

target surface.

The deposition apparatus is the Equipement Alliance Concept Cluster ACT200

(BHT, CEA Leti), which is composed of three deposition chambers. The co-

sputtering chamber is equipped with three targets of 100 mm of diameter that

can be used independently from each other or in co-sputtering to deposit different

compounds, and the substrate is a 200 mm Si wafer. The film uniformity is

improved by a satellite and sweeping motion of the target. The deposition is

done under Ar atmosphere at a pressure of 5 10−3 bar. The two other deposition

chambers are used to deposit electrodes and capping materials such as Ti, TiN

or Ta, while all the phase change thin films are deposited in the co-pulverization

chamber. The C-doped GeTe samples studied in Chapter 2 have been deposited

into the co-sputtering chamber by using two targets, one of GeTe and another

one of C. The N-doped GeTe samples have been deposited by pulverization of

a GeTe target in an Ar+N2 atmosphere. The film thickness is controlled by the

deposition time and the power applied to the targets. The thickness uniformity

is checked after deposition by 9 points measurements along the wafer diameter

and the films compositions are checked through Rutherford Back Scattering

(RBS).

138

List of Figures

1.1 Basic principle of the phase transformation [8]. PC materials can switch

reversibly between an amorphous state, corresponding to the logical level ’0’

or RESET, and a crystalline state corresponding to the logical level ’1’ or

SET. The SET operation consists in programming the cell into the SET state,

while the RESET operation consists in programming the cell into the RESET

state. To obtain the amorphous phase the PC material must be annealed

above its melting temperature and then rapidly cooled down. To obtain

the crystalline phase the material must be annealed above its crystallization

temperature Tx . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

1.2 Evidence of the different optical properties of the PC material Ge2Sb2Te5 in

the amorphous and crystalline phases. The reflectivity of GST is reported

as a function of temperature starting from an initially amorphous sample.

The amorphous phase is characterized by a low reflectivity value compared

to the one of the crystalline phase. On the graph it is easy to identify the

crystallization temperature at which the phase transformation occurs. . . . 14

1.3 Schematic representation of the lance-like structure of a PCM cell device.

The PC material is interfaced with a top electrode and a bottom electrode

(heater). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

1.4 Current pulses for the programming operation of the cell. RESET pulse (a)

SETMIN pulse (b) and SET pulse (c). . . . . . . . . . . . . . . . . . 16

1.5 I-V characteristic of a PCM cell in the crystalline and amorphous states (from

Ref.[9]). The I-V characteristic of the amorphous state present a snap-back

in correspondence of a threshold voltage that is not present in the crystalline

I-V curve. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

139

LIST OF FIGURES LIST OF FIGURES

1.6 Time-temperature-transformation (TTT) diagram for a PC material taken

from Reference [8]. The phase transformation of a fixed volume of PC mate-

rial is reported depending on the time spent at a certain temperature. The

two orange lines on the graph indicate two different constant rate quenching

processes while the two purple lines indicate two annealing processes starting

at room temperature. . . . . . . . . . . . . . . . . . . . . . . . . . 19

1.7 Evolution of ∆Gcluster(r) as a function of r corresponding to Eq. 1.2, taken

from Chapter 7 of Reference [4]. The curve exhibit a maximum for the r = rc

(critical radius) that corresponds to the critical work for cluster formation

∆Gc. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

1.8 Model for the heterogeneous nucleation taken from Chapter 7 of Reference

[4]. The crystalline cluster is a spherical cap which correspond to the ex-

posed part of a sphere of radius r. In the schematic picture are also re-

ported the wetting angle θ and the crystal-substrate, amorphous-substrate

and amorphous-crystal interfacial energies (respectively σcs, σls and σlc). . 24

1.9 PC materials reported on the ternary Ge:Sb:Te phase diagram, with the

GeTe− Sb2Te3 pseudo-binary line put in evidence (taken from Ref. [8]). . 28

1.10 Structure of GST in its crystalline metastable phase. One sublattice is oc-

cupied by Te atoms (light blue) while the other is randomly occupied by

Ge or Sb atoms (dark blue) or vacancies (around 20% ). The cubic lattice

parameter is 6.03 A [26]. . . . . . . . . . . . . . . . . . . . . . . . . 30

1.11 Structure of crystalline GeTe in its rhombohedral phase. The structure can

be described as a rocksalt-like structure, distorted by a relative shift of the

sublattices along the [111] direction. It is characterized by long (3.127 A)

and short (2.87 A) Ge-Te bonds shown respectively in white and green. . . 31

2.1 Low field cell resistance as a function of progam current for GST and GeTe

cells for various programming pulse times [24]. It can be noted that the SET

operation for the GeTe cell is faster and the difference in the resistance of

the amorphous and crystalline phases is higher. . . . . . . . . . . . . . 37

140


2.2 Calculation of the activation energy EA by interpolation of the fail times as

a function of 1/kT . In order to obtain the fail time, a PCM cell is written

in the RESET state and the fail time is defined as the time at which the

resistance of the cell is reduced by one half. . . . . . . . . . . . . . . . 38

2.3 Reflectivity measurements of C and N doped GeTe films (150 nm thick)

[45]. In both cases Tx increases with increasing doping concentration and

the effect is stronger for C doping. . . . . . . . . . . . . . . . . . . . 39

2.4 Activation energy (left) calculated for undoped and C-doped GeTe and low

electric field resistance as a function of the programming current (right) for

a GST, undoped GeTe and C-doped GeTe cell [44]. The activation energy

increases and the RESET current decreases with doping. . . . . . . . . . 39

2.5 Schematic representation of an X-ray incident beam scattered by a point-like

sample. The incident wavevector is k0, the scattered wavevector is kf and

the momentum transfer is Q = k0 − kf . . . . . . . . . . . . . . . . . . 41

2.6 Measured (a) S (Q) and (b) g (r) for undoped amorphous GeTe. It can be

noted that S (Q) tends to 1 for high Q. . . . . . . . . . . . . . . . . . 48

2.7 Comparison between the measured g(r) of amorphous (blue) and crystalline

(red) undoped GeTe. . . . . . . . . . . . . . . . . . . . . . . . . . . 49

2.8 Measured g (r) for (a) undoped GeTe and GeTeC (C=9.6% and 16.3%) and

(b) undoped GeTe and GeTeN (N=4% and 10%). In both cases, the first peak

is constant with doping while the intensity of the second peak of the undoped

sample decreases with increasing doping contents. A new peak appears at

around 3.5 A in the doped samples. These effects increase as a function of

doping and are stronger in the GeTeN case. . . . . . . . . . . . . . . . 50

2.9 Comparison between measured and calculated g (r) for undoped GeTe, GeTeC

(C=16.3% in the experiment and C=15% in the simulation) and GeTeN

(N=10% both in the experiment and in the simulation). Even if an already

known shift between peaks positions can be observed, the evolution of the

simulated and measured pair distribution functions with doping are in good

agreement. The effect of N-doping is stronger in the calculated g (r) than in

the measured one. . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

141


2.10 Partial pair distribution functions for (a) Ge-Ge, Te-Te, and Ge-Te pairs in

doped and undoped samples and (b) pairs involving C or N. Curves are shifted

for clarity. A new peak appears in the range 3.1-3.5 A in the Ge-Ge partial

pair distribution function for both C doped and N doped samples, while it

is absent in the undoped sample. A difference between partial contributions

involving C and the ones invloving N is the absence of Te-N bonds at small

distances (less than 3.2 A). . . . . . . . . . . . . . . . . . . . . . . . 53

2.11 Snapshot of the final state of the simulation box for GeTeC. Ge atoms are

represented in pink, Te atoms in light blue and C atoms in red. The in-

spection of this box combined with a bond angle analysis around C atoms,

reveals the presence of a mixture of tetrahedral (C − TeGe3, C − Ge4 and

C − Ge2Te2), triangular (C − C − Ge2 and C − C − GeTe), and linear (C

chains) bonds. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

2.12 Summary of the carbon environments in the C-doped GeTe sample. C −

TeGe3, C−Ge4 and C−Ge2Te2 tetrahedra can be found, as well as C−C−Ge2

and C− C−GeTe triangular environments. . . . . . . . . . . . . . . . 56

2.13 Snapshot of the final state of the simulation box for GeTeN. Ge atoms are

represented in pink, Te atoms in light blue and N atoms in green. N − Ge3

pyramidal environments, N−Ge4 tetrahedral environments and N2 molecules

can be found. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

2.14 Summary of the nitrogen environments found in the N-doped GeTe sample.

N−Ge4 tetrahedra, N−Ge3 pyramids and N2 molecules have been observed. 58

3.1 Resistivity as a function of time at room temperature for thin films of GST of

different thicknesses pre-annealed at 143.5C (from Ref.[67]). The incubation

time τ , defined as the time elapsed before the onset of crystallization, and the

transition time from the highest to lowest resistivity increase with decreasing

film thickness, meaning that the crystallization speed is reduced for small

thicknesses. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

3.2 Model used Ref.[71] to interpret the thickness-dependent variation of Tx .

A cylindric crystalline nucleus is embedded in the amorphous phase, sand-

wiched between two oxide interfaces. . . . . . . . . . . . . . . . . . . 66

142


3.3 Crystallization temperature Tx as a function of film thickness for various PC

materials: GST, N-doped GST (NGST), Ge15Sb85 (GeSb), Sb2Te and Ag-

and In-doped Sb2Te (AIST) deposited on Si and capped with Al2O3 , fitted

using Eq.3.1, as presented in Ref.[70]. . . . . . . . . . . . . . . . . . . 69

3.4 Sheet resistance of multilayered films of GST/SiO2 as a function of annealing

temperature, as reported in Ref.[78]. The label M25, M10 and M5 indicate

different bilayer thicknesses (M5 corresponds to the thinnest sample, M25 the

thickest one). The dotted lines correspond to ex situ annealing temperatures

used for further analysis in Ref.[78]. . . . . . . . . . . . . . . . . . . . 72

3.5 Scanning Electron Microscopy (SEM) image of as-grown GST nanowires from

Ref.[81]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

3.6 Measured values for the (a) recrystallization time at fixed temperature, (b)

nucleation rate and (c) activation energy as a function of nanowires diameter

as reported in Ref.[82]. . . . . . . . . . . . . . . . . . . . . . . . . . 74

3.7 Schematic drawing of the apparatus used to deposit the nanocluster sam-

ples as long as the thin film samples used for comparison. The deposition

procedure is briefly illustrated. . . . . . . . . . . . . . . . . . . . . . 78

3.8 (a) TOF size distribution which shows that the nanoclusters have an average

diameter of 5.7 nm with a narrow size distribution (± 1 nm at half maxi-

mum) (b) TEM images of GST as-deposited clusters which indicate that the

particles are spherical and amorphous. These images have been made on

a low density dedicated sample and the clusters state and shape have been

checked over clusters with the largest diameter. (By courtesy of M. Audier,

LMGP CNRS, Grenoble INP-Minatec . . . . . . . . . . . . . . . . . . 79

3.9 Scanning Electron Microscopy (SEM) image of GST deposited clusters. The

red circle has a diameter of 20 nm. No trace of particles coalescence can be

seen. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

3.10 (a) GST clusters sample: 4 layers of clusters capped with Al2O3 are deposited

on an Al2O3 substrate and capped again with Al2O3 . The average distance

between clusters in a layer is about 2 cluster diameters. (b) GST film sample:

10 nm thick film of GST sandwiched between two 10 nm thin Al2O3 films.

Both the clusters and film samples are deposited on a substrate of Si. . . . 81

143


3.11 (a) X-ray 2D diffraction images for 200C ex situ annealed GST thin film.

(b) Same measurements for 200C ex situ annealed clusters. In both cases,

the 2D image of the blank sample (Si+Al2O3 ) has been subtracted. . . . 83

3.12 X-ray diffraction spectra at room temperature for as-deposited and 200C ex

situ annealed GST film, after background subtraction, with curves shifted for

clarity. Arrows indicate bulk GST fcc peak positions calculated assuming the

lattice parameter of GST a=6.0117 A reported in Ref.[25]. . . . . . . . . 84

3.13 X-ray diffraction spectra at room temperature for as-deposited and 200C ex

situ annealed GST clusters after background subtraction. Curves are shifted

for clarity. Arrows indicate bulk GST fcc peak positions calculated assuming

the lattice parameter of GST a=6.0117 A reported in Ref.[25]. . . . . . . 85

3.14 (220) diffraction peak for in situ annealed GST thin film at different temper-

atures. The dotted line is the peak, measured at room temperature, of the

thin film annealed ex situ at 200C (shifted for clarity). The arrow indicates

the calculated bulk GST peak position. . . . . . . . . . . . . . . . . . 87

3.15 X-ray diffraction spectra for in situ annealed GST clusters at different tem-

peratures. (a) (200) diffraction peak and (b) (220) diffraction peak. Curves

are evenly shifted to ease viewing. Dotted lines indicate the peak position,

measured at room temperature, of the 200C ex situ annealed clusters. . . 88

3.16 Normalized integrated intensities for (220) and (200) diffraction peaks for

GST clusters and for GST film as a function of temperature. The normalized

integrated intensities have been obtained through the relation Inorm = I/Imax

where I is the measured integrated intensity at a given temperature and

Imax is the maximum value of the integrated intensity (which correspond

to complete crystallization). The dotted lines indicate the crystallization

temperatures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

144


3.17 The Al2O3 matrix surrounding the cluster forces its volume to remain equal

to the one of the amorphous phase even after crystallization. The cluster in its

amorphous phase occupies a certain volume (a). When crystallization occurs,

the cluster volume tends to reduce of around 5% (b), but the embedding

Al2O3 matrix exerts a tensile strain over the cluster (c) thus forcing the

cluster to keep the volume corresponding to the amorphous phase, with the

effect of increasing the lattice parameter of the crystalline cluster (d). . . . 91

4.1 Structure of PC material thin films samples used in order to investigate the

interface effect on crystallization. The PC material can be either GeTe or

GST, of various thicknesses, sandwiched between (a) SiO2 , (b) TiN or (c)

Ta. All the samples have been deposited by sputtering as described in B. . 97

4.2 Crystalline fraction as a function of temperature obtained from reflectivity

measurements for (a) GeTe and (b) GST 100 nm thin films sandwiched be-

tween TiN, Ta or SiO2 heated at 10C /min. For both GeTe and GST thin

films the amorphous to crystalline transition occurs at a higher temperature

when the film is sandwiched between Ta. . . . . . . . . . . . . . . . . 98

4.3 Crystalline fraction as a function of temperature from reflectivity measure-

ments for (a) GeTe 30 nm and (b) GeTe 10 nm thin films sandwiched between

TiN, Ta or SiO2 , heated at 10C /min. For 30 nm thick GeTe films the amor-

phous to crystalline phase transition occurs clearly at a higher temperature

when the PC material is sandwiched between Ta. For 10 nm thick films of

GeTe the measurement becomes difficult for samples interfaced with TiN and

Ta, while Tx can be still easily identified for the SiO2 interfaced sample. . 101

4.4 Kissinger plot for GST 100 nm thin films sandwiched between Ta or SiO2 .

The absolute value of the line slope corresponds to the activation energy EA .

The points in graph have been calculated from the Tx value obtained for four

different heating rates r as reported in Table 4.3 . . . . . . . . . . . . . 102

4.5 Schematic representation of the experimental geometry of the XRD experi-

ment, where θ is the incident beam angle and Ψ is the tilting angle of the

sample. A detailed description of the experimental setup is provided in ap-

pendix A.2.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

145


4.6 Diffracted intensity as a function of 2θ for Ψ =0of GeTe 100 nm thin films

sandwiched between TiN, Ta or SiO2 measured at 100C after annealing at

300C with a heating rate of about 0.11C /min. The vertical lines cor-

respond to the calculated position of Bragg peaks for rhombohedral GeTe

(hexagonal indexation) [25]. No difference in the diffraction spectra can be

observed between the Ta and TiN interfaced samples, while no peaks are

visible for the sample sandwiched in SiO2 . The intense peak around 33 is

due to Ta. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

4.7 012 Bragg peak for 100 nm thick GeTe films interfaced with SiO2 , TiN

and Ta measured at 100C after annealing at 230C (heating rate of about

0.11C /min) for various tilting angles Ψ of the samples. The Ta and TiN in-

terfaced samples show a weak texture while the sample sandwiched in SiO2 is

strongly textured, with a maximum peak intensity for Ψ=40 and no inten-

sity for Ψ=0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

4.8 Evolution of the GeTe 012 Bragg peak as a function of temperature observed

for Ψ=40 (heating rate of about 0.11C /min) for the 100nm thick GeTe

films interfaced with SiO2 , TiN and Ta. For each sample the thickest line

in the graph corresponds to the first temperature at which the Bragg peak

becomes visible. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

4.9 Evolution of the crystalline fraction as a function of temperature (heating

rate of about 0.11C /min) for the 100nm thick GeTe films interfaced with

SiO2 , TiN and Ta as obtained from the 012 GeTe Bragg peak area measured

by XRD through in situ annealing for a tilting angle Ψ=40. . . . . . . . 108

4.10 012 Bragg peak area as a function of temperature (heating rate of 0.11C /min)

and of the sample tilting angle Ψ for 100 nm GeTe films sandwiched between

TiN, Ta or SiO2 . The points on the descending temperature ramp are also

shown, and no evolution occurs during the cooling down process. . . . . . 109

4.11 Grain size calculated by Sherrer analysis on the 012 Bragg peak of GeTe 100

nm thin films interfaced with SiO2 , TiN or Ta. The error bar is about ±

5 nm. The dashed lines indicate the final dimensions of grains, measured at

100C after annealing, and are reported in Table 4.5. . . . . . . . . . . . 111

146


4.12 3D images of the diffracted rings obtained for GeTe 100 nm thin films an-

nealed at 400C for 15 minutes and interfaced with (a) SiO2 (b) TiN (c)

Ta. The SiO2 interfaced film is strongly textured while only a faint texture

is visible for the TiN interfaced film and the GeTe rings are isotropic for

the Ta interfaced film. The strongly textured rings that can be seen in (c)

correspond to Ta. . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

4.13 Diffracted intensity as a function of 2θ for the GeTe 100 nm thin film inter-

faced with SiO2 , measured at 100C after annealing at 300C (heating rate

around 0.11C /min) for various tilting angles Ψ. The intensity of the 012

Bragg peak is maximum at around Ψ=40, as already reported in Fig.4.7,

while the 101, 202, 104 and 110 peaks exhibit the highest intensity for Ψ=20.115

4.14 Secondary Ion Mass Spectrometry (SIMS) measurements performed on GeTe

and GST 30 nm thin films sandwiched with Ta. By definition, the interface

for each element can be placed in the point at which half of the signal in-

tensity is lost, corresponding to the vertical lines in the figure. From those

measurements the diffusion of Ta inside the GeTe and GST layers is extremely

low. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

4.15 Possible models of crystallization for (a) SiO2 and (b) Ta interfaced GeTe

thin films. Different colors correspond to different orientation of the grains.

In the case of SiO2 interface the crystallization begins with heterogeneous

nucleation at the energetically favorable interface and the nuclei grow with

a preferred orientation before the homogeneous nucleation starts. The final

result are bigger grains with a preferred orientation. In the case of Ta inter-

faced PC thin film, the heterogeneous nucleation at the interfaces is somehow

suppressed, thus the crystallization is driven by the homogeneous nucleation

that starts later respect the heterogeneous one, leading to a weak texture. . 118

147


A.1 Evidence of the different optical properties of the PC material Ge2Sb2Te5 in

the amorphous and crystalline phases. The reflectivity of GST is reported

as a function of temperature starting from an initially amorphous sample.

The amorphous phase is characterized by a low reflectivity value compared

to the one of the crystalline phase. On the graph it is easy to identify the

crystallization temperature at which the phase transformation occurs. . . . 126

A.2 Schematic representation of the reflectometer used for reflectivity measure-

ments. The laser beam is directed onto a birefringent filter and divided in

two beams, and one of them is directed to the sample. The direct and re-

flected beams are collected by a photodetector and processed to obtain the

measured signal. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

A.3 Schematic representation of the geometry used for XRD analysis in labora-

tory. The experiment is performed in a θ-2θ configuration (θ1 = θ2) and the

tilting angle Ψ allows to measure the sample texture. . . . . . . . . . . . 128

A.4 Picture and schematic representation of the experimental setup used at the

synchrotron SOLEIL. The scattered transmitted beam is collected by an

image plate detector placed at a distance D≈ 21cm from the sample, which

is the minimum allowed distance in this configuration. Thus, in order to

obtain a high value of Q, the center of the image does not correspond to the

center of the detector. . . . . . . . . . . . . . . . . . . . . . . . . . 130

A.5 Image acquired on the capillary containing amorphous GeTe. Only the pixels

contained in the vertical triangular sector shown on the figure have been

selected for integration. The pixels that appear as a white spot at the center

of the image plate are damaged and must be avoided. . . . . . . . . . . 132

A.6 Schematic representation of the experimental setup used at the synchrotron

ESRF. The diffracted beam is collected by a CCD camera at a distance D

from the sample of around 20 cm. The camera and the sample are tilted of

an angle θ2 and θ1 with respect to the incident beam, respectively. If θ1 = θ2,

the configuration is in a strict θ −2θ geometry only at the center of the camera.133

148

List of Tables

1.1 Properties that characterize PC materials [7]. . . . . . . . . . . . . . . 27

1.2 Comparison between the main properties of Ge2Sb2Te5 (GST) and GeTe. . 29

2.1 Crystallization temperatures Tx of C and N doped GeTe films (150 nm thick),

taken as the midpoint of the rising steps of the reflectivity curves reported

in Fig.2.3 [45]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

2.2 Measured mass densities and atomic number densities for Ge52Te48, undoped

and doped with carbon or nitrogen, expressed in g/cm3 and atoms/A3, re-

spectively. The mass densities have been measured by X-ray reflectivity (XRR). 46

3.1 Crystallization temperatures as a function of heating rate and GST film

thickness from Ref.[67]. . . . . . . . . . . . . . . . . . . . . . . . . . 67

3.2 Peak positions and relative intensities for the fcc GST phase as expected from

a powder pattern. They have been estimated in a θ −2θ geometry at the

actual experimental wavelength considering the lattice parameter a=6.0117

A reported in Ref [25]. . . . . . . . . . . . . . . . . . . . . . . . . . 86

4.1 Crystallization temperatures Tx of GeTe and GST 100 nm thick films sand-

wiched between TiN, Ta or SiO2 as obtained from the reflectivity measure-

ments of Fig. 4.2 for a heating rate of 10C /min. . . . . . . . . . . . . 99

4.2 Crystallization temperatures Tx of GeTe films 30 nm and 10 nm thick sand-

wiched between TiN, Ta or SiO2 as obtained from the reflectivity measure-

ments of Fig. 4.3 for a heating rate of 10C /min. . . . . . . . . . . . . 100

4.3 Crystallization temperature Tx obtained from reflectivity measurements for

different heating rates for GeTe 100 nm thin films sandwiched between Ta or

SiO2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

149

LIST OF TABLES LIST OF TABLES

4.4 Crystallization temperatures Tx of 100 nm thick GeTe films sandwiched be-

tween TiN, Ta or SiO2 , obtained as the temperatures corresponding to the

midpoints of the rise steps of the Bragg peaks areas as a function of tempe-

rature reported in Fig. 4.9 (heating rate of 0.11C /min). . . . . . . . . 108

4.5 Final mean grain sizes measured at 100C after annealing for different values

of the tilting angle Ψ for 100nm thick GeTe films. . . . . . . . . . . . . 110

4.6 Summary of the different characteristics observed for GeTe thin films encap-

sulated in SiO2 , TiN or Ta. The SiO2 and Ta interfaced samples present the

most relevant differences in their properties, while the TiN interfaced sample

can be considered as an intermediate situation. . . . . . . . . . . . . . 114

150

Bibliography

[1] G.E. Moore. Cramming more components onto integrated circuits. Elec-

tronics, 38(8):114–117, 1965.

[2] G. W. Burr, M. J. Breitwisch, M. Franceschini, D. Garetto, K. Gopalakr-

ishnan, B. Jackson, B. Kurdi, C. Lam, L. A. Lastras, A. Padilla, B. Rajen-

dran, S. Raoux, and R. S. Shenoy. Phase change memory technology. Jour-

nal of Vacuum Science & Technology B: Microelectronics and Nanometer

Structures, 28(2):223, 2010.

[3] S. Ovshinsky. Reversible electrical switching phenomena in disordered

structures. Physical Review Letters, 21(20):1450–1453, November 1968.

[4] S. Raoux and M. Wuttig. Phase change materials: science and applica-

tions. Springer Verlag, 2008.

[5] N. Yamada, E. Ohno, N. Akahira, K. Nishiuchi, K. Nagata, and M. Takao.

High speed overwritable phase change optical disk material. Japanese

Journal of Applied Physics, 26S4(Supplement 26-4):61–66, 1987.

[6] N. Yamada, E. Ohno, K. Nishiuchi, N. Akahira, and M. Takao. Rapid-

phase transitions of GeTe−Sb2Te3 pseudobinary amorphous thin films for

an optical disk memory. Journal of Applied Physics, 69(5):2849, 1991.

[7] M. Wuttig and N. Yamada. Phase-change materials for rewriteable data

storage. Nature Materials, 6(11):824–832, November 2007.

[8] D. Lencer, M. Salinga, and M. Wuttig. Design rules for phase-change

materials in data storage applications. Advanced Materials, 23(18):2030–

2058, May 2011.

151

BIBLIOGRAPHY BIBLIOGRAPHY

[9] G. Betti Beneventi. Characterization and modeling of phase-change mem-

ories. PhD Thesis, 2011.

[10] D. Ielmini and Y. Zhang. Analytical model for subthreshold conduction

and threshold switching in chalcogenide-based memory devices. Journal

of Applied Physics, 102(5):054517–054517–13, September 2007.

[11] D. Ielmini. Threshold switching mechanism by high-field energy gain

in the hopping transport of chalcogenide glasses. Physical Review B,

78(3):035308, 2008.

[12] J. A. Kalb. Crystallization kinetics in antimony and tellurium alloys used

for phase change recording. Doctoral thesis, RWTH Aachen University,

2006.

[13] J. W. Christian. The Theory of Transformations in Metals and Alloys.

Pergamon Press, 1975.

[14] K. F. Kelton, H. Ehrenreich, and D. Turnbull. Crystal nucleation in liquids

and glasses. In Solid State Physics, volume Volume 45, pages 75–177.

Academic Press, 1991.

[15] O. Penrose. The Becker-Dring equations for the kinetics of phase transi-

tions. August, 2001.

[16] D. Turnbull and J. C. Fisher. Rate of nucleation in condensed systems.

The Journal of Chemical Physics, 17(1):71, 1949.

[17] J. Willard Gibbs. The scientific papers of J. Willard Gibbs. Dover, New

York, 1961.

[18] D. T. Wu. Nucleation theory. Solid State Physics - Advances in Research

and Applications, Vol 50, 50:37–187, 1997.

[19] H. E. Kissinger. Reaction kinetics in differential thermal analysis. Ana-

lytical chemistry, 29(11):1702–1706, 1957.

152


[20] S. Senkader and D. C. Wright. Models for phase-change of Ge2Sb2Te5

in optical and electrical memory devices. Journal of Applied Physics,

95(2):504, 2004.

[21] H.Y. Cheng, T.H. Hsu, S. Raoux, J.Y. Wu, P.Y. Du, M. Breitwisch,

Y. Zhu, E.K. Lai, E. Joseph, S. Mittal, R. Cheek, A. Schrott, S.C. Lai,

H.L. Lung, and C. Lam. A high performance phase change memory with

fast switching speed and high temperature retention by engineering the

GexSbyTez phase change material. In Electron Devices Meeting (IEDM),

2011 IEEE International, pages 3.4.1 –3.4.4, December 2011.

[22] A. Fantini, L. Perniola, M. Armand, J.F. Nodin, V. Sousa, A. Persico,

J. Cluzel, C. Jahan, S. Maitrejean, S. Lhostis, A. Roule, C. Dressler,

G. Reimbold, B. De Salvo, P. Mazoyer, D. Bensahel, and F. Boulanger.

Comparative assessment of GST and GeTe materials for application to

embedded phase-change memory devices. In Memory Workshop, 2009.

IMW ’09. IEEE International, pages 1–2, 2009.

[23] E. Gourvest, B. Pelissier, C. Valle, A. Roule, S. Lhostis, and S. Maitrejean.

Impact of oxidation on Ge2Sb2Te5 and GeTe phase-change properties.

Journal of The Electrochemical Society, 159(4):H373H377, 2012.

[24] L. Perniola, V. Sousa, A. Fantini, E. Arbaoui, A. Bastard, M. Armand,

A. Fargeix, C. Jahan, J.-F. Nodin, A. Persico, D. Blachier, A. Toffoli,

S. Loubriat, E. Gourvest, G. Betti Beneventi, H. Feldis, S. Maitrejean,

S. Lhostis, A. Roule, O. Cueto, G. Reimbold, L. Poupinet, T. Billon,

B. De Salvo, D. Bensahel, P. Mazoyer, R. Annunziata, P. Zuliani, and

F. Boulanger. Electrical behavior of phase-change memory cells based on

GeTe. IEEE Electron Device Letters, 31(5):488 –490, May 2010.

[25] T. Nonaka, G. Ohbayashi, Y. Toriumi, Y. Mori, and H. Hashimoto. Crystal

structure of GeTe and Ge2Sb2Te5 meta-stable phase. Thin Solid Films,

370(1):258–261, 2000.

153


[26] N. Yamada and T. Matsunaga. Structure of laser-crystallized Ge2Sb2 +

xTe5 sputtered thin films for use in optical memory. Journal of Applied

Physics, 88(12):7020–7028, December 2000.

[27] Shinji Kohara, Kenichi Kato, Shigeru Kimura, Hitoshi Tanaka, Takeshi

Usuki, Kentaro Suzuya, Hiroshi Tanaka, Yutaka Moritomo, Toshiyuki

Matsunaga, Noboru Yamada, Yoshihito Tanaka, Hiroyoshi Suematsu, and

Masaki Takata. Structural basis for the fast phase change of ge[sub

2]Sb[sub 2]Te[sub 5]: Ring statistics analogy between the crystal and amor-

phous states. Applied Physics Letters, 89(20):201910, 2006.

[28] J. Akola and R. Jones. Structural phase transitions on the nanoscale:

The crucial pattern in the phase-change materials Ge2Sb2Te5 and GeTe.

Physical Review B, 76(23), December 2007.

[29] A. V. Kolobov, P. Fons, A. I. Frenkel, A. L. Ankudinov, J. Tominaga,

and T. Uruga. Understanding the phase-change mechanism of rewritable

optical media. Nature Materials, 3(10):703–708, September 2004.

[30] S. Hosokawa, T. Ozaki, K. Hayashi, N. Happo, M. Fujiwara, K. Horii,

P. Fons, A. V. Kolobov, and J. Tominaga. Existence of tetrahedral site

symmetry about Ge atoms in a single-crystal film of Ge2Sb2Te5 found

by x-ray fluorescence holography. Applied Physics Letters, 90(13):131913,

2007.

[31] K. S. Andrikopoulos, S. N. Yannopoulos, G. A. Voyiatzis, A. V. Kolobov,

M. Ribes, and J. Tominaga. Raman scattering study of the a-GeTe struc-

ture and possible mechanism for the amorphous to crystal transition. Jour-

nal of Physics: Condensed Matter, 18(3):965–979, January 2006.

[32] K.S. Andrikopoulos, S.N. Yannopoulos, A.V. Kolobov, P. Fons, and

J. Tominaga. Raman scattering study of GeTe and Ge2Sb2Te5 phase-

change materials. Journal of Physics and Chemistry of Solids, 68(5-

6):1074–1078, May 2007.

154


[33] J. Akola and R. Jones. Binary alloys of Ge and Te: Order, voids, and the

eutectic composition. Physical Review Letters, 100(20), May 2008.

[34] R. Kojima, S. Okabayashi, T. Kashihara, K. Horai, T. Matsunaga,

E. Ohno, N. Yamada, and T. Ohta. Nitrogen doping effect on phase

change optical disks. Japanese Journal of Applied Physics, 37(Part 1, No.

4B):2098–2103, 1998.

[35] S. Y. Kim, S. J. Kim, H. Seo, and M. R. Kim. Complex refractive indices of

GeSbTe-Alloy thin films: Effect of nitrogen doping and wavelength depen-

dence. Japanese Journal of Applied Physics, 38(Part 1, No. 3B):1713–1714,

1999.

[36] S. Privitera, E. Rimini, and R. Zonca. Amorphous-to-crystal transition of

nitrogen- and oxygen-doped Ge2Sb2Te5 films studied by in situ resistance

measurements. Applied Physics Letters, 85(15):3044, 2004.

[37] K. Kim, J. Park, J. Chung, S. A. Song, M. Jung, Y. M. Lee, H-J Shin,

B. Kuh, Y. Ha, and J-S Noh. Observation of molecular nitrogen in n-doped

Ge2Sb2Te5. Applied Physics Letters, 89(24):243520–243520–3, December

2006.

[38] L. W. Fang, Z. Zhang, R. Zhao, J. Pan, M. Li, L. Shi, T. Chong, and

Y. Yeo. Fermi-level pinning and charge neutrality level in nitrogen-doped

Ge2Sb2Te5 : Characterization and application in phase change memory

devices. Journal of Applied Physics, 108(5):053708–053708–7, September

2010.

[39] H. Horii, J.H. Yi, J.H. Park, Y.H. Ha, I.G. Baek, S.O. Park, Y.N. Hwang,

S.H. Lee, Y.T. Kim, K.H. Lee, U.-In Chung, and J.T. Moon. A novel cell

technology using n-doped GeSbTe films for phase change RAM. In 2003

Symposium on VLSI Technology, 2003. Digest of Technical Papers, pages

177 – 178, June 2003.

[40] S.J. Ahn, Y.J. Song, C.W. Jeong, J.M. Shin, Y. Fai, Y.N. Hwang, S.H. Lee,

K.C. Ryoo, S.Y. Lee, J.H. Park, H. Horii, Y.H. Ha, J.H. Yi, B.J. Kuh,

155


G.H. Koh, G.T. Jeong, H.S. Jeong, Kinam Kim, and B.I. Ryu. Highly

manufacturable high density phase change memory of 64Mb and beyond.

In Electron Devices Meeting, 2004. IEDM Technical Digest. IEEE Inter-

national, pages 907 – 910, December 2004.

[41] A. Fantini, V. Sousa, L. Perniola, E. Gourvest, J.C. Bastien, S. Maitrejean,

S. Braga, N. Pashkov, A. Bastard, B. Hyot, A. Roule, A. Persico, H. Feldis,

C. Jahan, J.F. Nodin, D. Blachier, A. Toffoli, G. Reimbold, F. Fillot,

F. Pierre, R. Annunziata, D. Benshael, P. Mazoyer, C. Vallee, T. Billon,

J. Hazart, B. De Salvo, and F. Boulanger. N-doped GeTe as performance

booster for embedded phase-change memories. In Electron Devices Meeting

(IEDM), 2010 IEEE International, pages 29.1.1 –29.1.4, December 2010.

[42] K. Kim, S. Choi, J. Chung, J. Lee, and S. Heo. Changes in chemical and

structural properties of phase-change material GeTe with nitrogen doping

and annealing. Japanese Journal of Applied Physics, 49(6):061801, June

2010.

[43] C. Peng, L. Wu, F. Rao, Z. Song, X. Zhou, M. Zhu, B. Liu, D. Yao,

S. Feng, P. Yang, and J. Chu. Nitrogen incorporated GeTe phase change

thin film for high-temperature data retention and low-power application.

Scripta Materialia, 65(4):327–330, 2011.

[44] G. Betti Beneventi, L. Perniola, V. Sousa, E. Gourvest, S. Maitrejean, J.C.

Bastien, A. Bastard, B. Hyot, A. Fargeix, C. Jahan, J.F. Nodin, A. Per-

sico, A. Fantini, D. Blachier, A. Toffoli, S. Loubriat, A. Roule, S. Lhostis,

H. Feldis, G. Reimbold, T. Billon, B. De Salvo, L. Larcher, P. Pavan,

D. Bensahel, P. Mazoyer, R. Annunziata, P. Zuliani, and F. Boulanger.

Carbon-doped GeTe: a promising material for phase-change memories.

Solid-State Electronics, 6566(0):197–204, November 2011.

[45] P. Noe, J. Y. Raty, G. E. Ghezzi, F. Hippert, E. Souchier, L. Perniola,

V. Sousa, A. Persico, E Henaff, M. Tessaire, C. Bichara, and S. Maitrejean.

Effect of dopant incorporation in phase change material. E\PCOS 2012

Proceedings, 2012.

156


[46] G. E. Ghezzi, J. Y. Raty, S. Maitrejean, A. Roule, E. Elkaim, and F. Hip-

pert. Effect of carbon doping on the structure of amorphous GeTe phase

change material. Applied Physics Letters, 99(15):151906, 2011.

[47] D. A. Keen. A comparison of various commonly used correlation func-

tions for describing total scattering. Journal of Applied Crystallography,

34(2):172–177, April 2001.

[48] H. E. Fischer, A. C. Barnes, and P. S Salmon. Neutron and x-ray

diffraction studies of liquids and glasses. Reports on Progress in Physics,

69(1):233–299, January 2006.

[49] T. E. Faber and J. M. Ziman. A theory of the electrical properties of liquid

metals. Philosophical Magazine, 11(109):153–173, 1965.

[50] M. Micoulaut, J.-Y. Raty, C. Otjacques, and C. Bichara. Understanding

amorphous phase-change materials from the viewpoint of maxwell rigidity.

Physical Review B, 81(17), May 2010.

[51] J-Y. Raty, C. Otjacques, J-P. Gaspard, and C. Bichara. Amorphous struc-

ture and electronic properties of the ge1sb2te4 phase change material. Solid

State Sciences, 12(2):193–198, February 2010.

[52] J. Akola and R. Jones. Structure of amorphous Ge8Sb2Te11 GeTe-Sb2Te3

alloys and optical storage. Physical Review B, 79(13), April 2009.

[53] C. Bichara and J-Y. Raty. Temperature-induced density anomaly in te-

rich liquid germanium tellurides: p versus sp3 bonding. Physical Review

Letters, 95(26), December 2005.

[54] J. Akola, J. Larrucea, and R. Jones. Polymorphism in phase-change

materials: melt-quenched and as-deposited amorphous structures in

Ge2Sb2Te5 from density functional calculations. Physical Review B, 83(9),

March 2011.

157


[55] T. Chattopadhyay, J. X. Boucherle, and H. G. VonSchnering. Neutron

diffraction study on the structural phase transition in GeTe. Journal of

Physics C: Solid State Physics, 20(10):1431–1440, April 1987.

[56] Alexander V. Kolobov, Paul Fons, Berangere Hyot, Bernard Andre, Junji

Tominaga, Yusuke Tamenori, Hideki Yoshikawa, and Keisuke Kobayashi.

Local structure of nitrogen in n-doped amorphous and crystalline GeTe.

Applied Physics Letters, 100(6):061910, 2012.

[57] K. B. Borisenko, Y. Chen, D. J.H. Cockayne, S. A. Song, and H. S. Jeong.

Understanding atomic structures of amorphous c-doped Ge2Sb2Te5 phase-

change memory materials. Acta Materialia, 59(11):4335–4342, June 2011.

[58] E. Cho, Y. Youn, and S. Han. Enhanced amorphous stability of carbon-

doped Ge2Sb2Te5 : Ab initio investigation. Applied Physics Letters,

99(18):183501–183501–3, November 2011.

[59] Eunae Cho, Seungwu Han, Dohyung Kim, Hideki Horii, and Ho-Seok

Nam. Ab initio study on influence of dopants on crystalline and amor-

phous Ge2Sb2Te5. Journal of Applied Physics, 109(4):043705–043705–10,

February 2011.

[60] Y.M. Lee, H.J. Shin, S.J. Choi, J.H. Oh, H.S. Jeong, K. Kim, and M.-

C. Jung. Nitrogen contribution to N-doped GeTe (N: 8.4at.%) in the

structural phase transition. Current Applied Physics, 11(3):710–713, May

2011.

[61] J-Y. Raty, P. Noe, G. E. Ghezzi, S. Maitrejean, C. Bichara, and F. Hip-

pert. Vibrational properties and stabilization mechanism of the doped gete

phase change material amorphous phase. Submitted to Phys. Rev. Lett.,

().

[62] X. Zhou, L. Wu, Z. Song, F. Rao, M. Zhu, C. Peng, D. Yao, S. Song,

B. Liu, and S. Feng. Carbon-doped Ge2Sb2Te5 phase change material:

A candidate for high-density phase change memory application. Applied

Physics Letters, 101(14):142104–142104–4, October 2012.

158


[63] Xavier Biquard, Milos Krbal, Alexander V. Kolobov, Paul Fons, Robert E.

Simpson, Brangre Hyot, Bernard Andr, Junji Tominaga, and Tomoya

Uruga. Effect of doping on global and local order in crystalline GeTe.

Applied Physics Letters, 98(23):231907–231907–3, June 2011.

[64] B. Liu, Z. Song, T. Zhang, J. Xia, S. Feng, and B. Chen. Effect of n-

implantation on the structural and electrical characteristics of Ge2Sb2Te5

phase change film. Thin Solid Films, 478(12):49–55, 2005.

[65] G. E. Ghezzi, R. Morel, A. Brenac, N. Boudet, M. Audier, F. Fillot,

S. Maitrejean, and F. Hippert. Crystallization of Ge2Sb2Te5 nanometric

phase change material clusters made by gas-phase condensation. Applied

Physics Letters, 101(23):233113–233113–4, December 2012.

[66] N. Ohshima. Crystallization of germaniumantimonytellurium amorphous

thin film sandwiched between various dielectric protective films. Journal

of Applied Physics, 79(11):8357, 1996.

[67] X. Wei, S. Luping, Z. Chong, C. T.and Rong, and L.H. Koon. Thick-

ness dependent nano-crystallization in Ge2Sb2Te5 films and its effect on

devices. Japanese Journal of Applied Physics, 46(4B):2211–2214, April

2007.

[68] S. Privitera, S. Lombardo, C. Bongiorno, E. Rimini, and A. Pirovano.

Phase change mechanisms in Ge2Sb2Te5 . Journal of Applied Physics,

102(1):013516, 2007.

[69] J. Kalb, F. Spaepen, and M. Wuttig. Calorimetric measurements of phase

transformations in thin films of amorphous te alloys used for optical data

storage. Journal of Applied Physics, 93(5):2389, 2003.

[70] S. Raoux, Jean L. J.-S., and A. J. Kellock. Crystallization properties of

ultrathin phase change films. Journal of Applied Physics, 103(11):114310,

2008.

159


[71] M. Zacharias and P. Streitenberger. Crystallization of amorphous super-

lattices in the limit of ultrathin films with oxide interfaces. Physical Review

B, 62(12):8391, 2000.

[72] R. E. Simpson, M. Krbal, P. Fons, A. V. Kolobov, J. Tominaga, T. Uruga,

and H. Tanida. Toward the ultimate limit of phase change in Ge2Sb2Te5 .

Nano letters, 10(2):414419, 2009.

[73] W. D. Nix. Mechanical properties of thin films. Metallurgical Transactions

A, 20(11):2217–2245, November 1989.

[74] H.-Y. Cheng, S. Raoux, C.-W. Cheng, and J. L. Jordan-Sweet. The crys-

tallization behavior of stressed Ge2Sb2Te5 phase-change material. MRS

Spring Meeting 2012, 2012.

[75] F. Oki, Y. Ogawa, and Y. Fujiki. Effect of deposited metals on the crys-

tallization temperature of amorphous germanium film. Japanese Journal

of Applied Physics, 8:1056, 1969.

[76] H. Homma, I. K. Schuller, W. Sevenhans, and Y. Bruynseraede. Inter-

facially initiated crystallization in amorphous germanium films. Applied

Physics Letters, 50(10):594, 1987.

[77] N. Ohshima. Structural analysis and crystallization studies of germani-

umantimony tellurium sputtered films on different underlayers. 83(10),

1998.

[78] M. H. Jang, S. J. Park, D. H. Lim, M.-H. Cho, Y. K. Kim, H.-J. Yi,

and H. S. Kim. Structural stability and phase-change characteristics of

Ge2Sb2Te5 nano-multilayered films. Electrochemical and Solid-State Let-

ters, 12(4):H151, 2009.

[79] S. Meister, H. Peng, K. McIlwrath, K. Jarausch, X. F. Zhang, and Y. Cui.

Synthesis and characterization of phase-change nanowires. Nano letters,

6(7):15141517, 2006.

160


[80] D. Yu, J. Wu, Q. Gu, and H. Park. Germanium telluride nanowires and

nanohelices with memory-switching behavior. Journal of the American

Chemical Society, 128(25):8148–8149, 2006.

[81] S.-H. Lee, Y. Jung, and R. Agarwal. Highly scalable non-volatile and

ultra-low-power phase-change nanowire memory. Nature Nanotechnology,

2(10):626–630, September 2007.

[82] S.-H. Lee, Y. Jung, and R. Agarwal. Size-dependent surface-induced het-

erogeneous nucleation driven phase-change in Ge2Sb2Te5 nanowires. Nano

letters, 8(10):3303–3309, October 2008. PMID: 18778106.

[83] S. Raoux, C. T. Rettner, J. L. Jordan-Sweet, A. J. Kellock, T. Topuria,

P. M. Rice, and D. C. Miller. Direct observation of amorphous to crys-

talline phase transitions in nanoparticle arrays of phase change materials.

Journal of Applied Physics, 102(9):094305, 2007.

[84] Y. Zhang, S. Raoux, D. Krebs, L. E. Krupp, T. Topuria, M. A. Caldwell,

D. J. Milliron, A. Kellock, Philip M. Rice, J. L. Jordan-Sweet, and H.-S. P.

Wong. Phase change nanodots patterning using a self-assembled poly-

mer lithography and crystallization analysis. Journal of Applied Physics,

104(7):074312, 2008.

[85] H. S. Choi, K. S. Seol, K. Takeuchi, J. Fujita, and Y. Ohki. Synthesis of

size- and structure-controlled Ge2Sb2Te5 nanoparticles. Japanese Journal

of Applied Physics, 44(10), 2005.

[86] H. R. Yoon, W. Jo, E. H. Lee, J. H. Lee, M. Kim, K. Y. Lee, and Y. Khang.

Generation of phase-change Ge–Sb–Te nanoparticles by pulsed laser abla-

tion. Journal of non-crystalline solids, 351(43):3430–3434, 2005.

[87] G.-S. Park, J.-H. Kwon, M. Kim, H. R. Yoon, W. Jo, T. K. Kim, J.-M.

Zuo, and Y. Khang. Crystalline and amorphous structures of Ge–Sb–Te

nanoparticles. Journal of Applied Physics, 102(1):013524, 2007.

161


[88] I. U. Arachchige, R. Soriano, C. D. Malliakas, S. A. Ivanov, and M. G.

Kanatzidis. Amorphous and crystalline GeTe nanocrystals. Advanced

Functional Materials, 21(14):2737–2743, July 2011.

[89] M. A. Caldwell, S. Raoux, R. Y. Wang, H.-S. Philip Wong, and D. J. Mil-

liron. Synthesis and size-dependent crystallization of colloidal germanium

telluride nanoparticles. Journal of Materials Chemistry, 20(7):1285, 2010.

[90] M. J. Polking, H. Zheng, R. Ramesh, and A. P. Alivisatos. Controlled

synthesis and size-dependent polarization domain structure of colloidal

germanium telluride nanocrystals. Journal of the American Chemical So-

ciety, 2011.

[91] R. Morel, A. Brenac, P. Bayle-Guillemaud, C. Portemont, and F. La Rizza.

Growth and properties of cobalt clusters made by sputtering gas-

aggregation. European Physical Journal D, 24(1-3):287–290, June 2003.

[92] H.R. Yoon, W. Jo, E. Cho, S. Yoon, and M. Kim. Microstructure and

optical properties of phase-change GeSbTe nanoparticles grown by pulsed-

laser ablation. Journal of Non-Crystalline Solids, 352(3637):3757–3761,

October 2006.

[93] W. K. Njoroge, H.-W. Woltgens, and M. Wuttig. Density changes upon

crystallization of Ge2Sb2.04Te4.74 films. Journal of Vacuum Science & Tech-

nology A: Vacuum, Surfaces, and Films, 20(1):230, 2002.

[94] T. P. Leervad Pedersen, J. Kalb, W. K. Njoroge, D. Wamwangi, M. Wuttig,

and F. Spaepen. Mechanical stresses upon crystallization in phase change

materials. Applied Physics Letters, 79(22):3597, 2001.

[95] D.-M. Smilgies. Scherrer grain-size analysis adapted to grazing-incidence

scattering with area detectors. Journal of Applied Crystallography,

42(6):1030–1034, October 2009.

[96] S. Meister, D. T. Schoen, M. A. Topinka, A. M. Minor, and Y. Cui. Void

formation induced electrical switching in phase-change nanowires. Nano

Letters, 8(12):4562–4567, December 2008.

162


[97] V. Weidenhof, I. Friedrich, S. Ziegler, and M. Wuttig. Laser induced

crystallization of amorphous Ge2Sb2Te5 films. Journal of Applied Physics,

89(6):3168, 2001.

[98] L. H. Zhang, E. Johnson, and U. Dahmen. Observations of vacancy-

controlled decay of elastic strain caused by phase transformation of small

pb inclusions in al. Acta materialia, 53(13):36353642, 2005.

[99] K. J. Choi, S. M. Yoon, N. Y. Lee, S. Y. Lee, Y. S. Park, B. G. Yu, and

S. O. Ryu. The effect of antimony-doping on Ge2Sb2Te5 , a phase change

material. Thin Solid Films, 516(23):88108812, 2008.

[100] M. Wuttig, D. Lsebrink, D. Wamwangi, W. Wenic, M. Gillessen, and

R. Dronskowski. The role of vacancies and local distortions in the design

of new phase-change materials. Nature Materials, 6(2):122–128, December

2006.

[101] J.A. Kalb, M. Wuttig, and F. Spaepen. Calorimetric measurements of

structural relaxation and glass transition temperatures in sputtered films

of amorphous te alloys used for phase change recording. Journal of Mate-

rials Research, 22(03):748–754, March 2011.

[102] A. Marmier, K. Kohary, and D. C. Wright. Determination of the

anisotropic elastic properties of Ge1Sb2Te4. Applied Physics Letters,

98:231911, 2011.

[103] S. Raoux, B. Munoz, H.-Y. Cheng, and J. L. Jordan-Sweet. Phase tran-

sitions in Ge–Te phase change materials studied by time-resolved x-ray

diffraction. Applied Physics Letters, 95(14):143118, 2009.

[104] M. Chen, K. A. Rubin, and R. W. Barton. Compound materials for

reversible, phasechange optical data storage. Applied Physics Letters,

49(9):502–504, September 1986.

[105] C. V. Thompson and R. Carel. Stress and grain growth in thin films.

Journal of the Mechanics and Physics of Solids, 44(5):657673, 1996.

163


[106] B.J. Mrstik, H.L. Hughes, P. Gouker, R.K. Lawrence, and P.J. McMarr.

The role of nanoclusters in reducing hole trapping in ion implanted oxides.

IEEE Transactions on Nuclear Science, 50(6):1947 – 1953, December 2003.

[107] A.P. Hammersley. FIT2D: an introduction and overview. ESRF Internal

Report, ESRF97HA02T, 1997.

[108] X. Qiu, J. W. Thompson, and S. J. L. Billinge. PDFgetX2: a GUI-

driven program to obtain the pair distribution function from x-ray powder

diffraction data. Journal of Applied Crystallography, 37(4):678–678, July

2004.

164

Acknowledgments

First of all, I would like to thank with all my heart my Thesis Director

Francoise Hippert. With her exceptional energy and dedication to work, not to

mention her sense of humor, she helped and encouraged me constantly through

these three years like no-one else could do. I learnt much from her, and we had

such wonderful (even when hard!) times between synchrotrons, office work at

night and conferences that filled those three years with beautiful memories I

will never forget. My deep gratitude goes also to the co-Director of this thesis,

Sylvain Maitrejean, that was able to take care of me even if time was never

enough, always with an optimistic aptitude that can make any problem easier.

I would also like to thank with all my heart Frederic Fillot, who taught me so

much and with whom I had many wonderful conversations that I really treasure.

I wish we had more time to spend together.

A great thank goes to all the PCM team of CEA Leti. First of all to the project

leader Veronique Sousa and to Pierre Noe, not only for their competence in work,

patience, dynamism and optimism, but also for being the best companions that

one can desire when night never comes in Finland. I thank Luca Perniola, great

worker and solid reference for all the italians PhD students, for the fruitful

discussions and for the good laughs. Thanks to Olga Cueto for helping me

with learning COMSOL, to Alain Persico, Carine Jahan, Jean-Francois Nodin,

Christope Vallee, Philippe Michallon, and to everyone that I could have forgotten

to add. Thanks to Anne Roule, Mathieu Petit and Ewen Henaff for the thin films

samples depositions. I deeply thank Robert Morel and Ariel Brenac for their

165

Acknowledgments

wonderful clusters deposition and for the good time we had while characterizing

them, as well as for their enormous knowledge and competence that helped me so

much. I am particularly grateful to Jean-Yves Raty, who can find a theoretical

explanation for all our experimental evidences even before they are found (really

impressive!). Thanks to Jean-Paul Simon, Nathalie Boudet and Jean-Francois

Berard, that were our precious contact on the BM02 D2NT beamline at the

ESRF, and to Erik Elkaim that supported us so much at SOLEIL.

My gratitude goes to all the PhD students of the laboratory, old and new. I

would like to thank in particular Audrey who is one of the kindest persons I

have ever met, and my wonderful cobureau Kavita, Sylvia, Raul and Bilel with

whom I divided for almost two years a small office room that felt like home.

I deeply thank all the PhD students of the PCM group, each of whom placed

its brick in building the PCM project: Giovanni e Stefania, Emmanuel, Jean

Claude, Eddie, Gabriele, Quentin, Manan, the sweet Sarra and everyone else.

I hope I did not forget anyone, and I apologize if I did so. Thank you all, I wish

you all the best.

I would like to make some more personal acknowledgments and the best

thing to do is to write them in the own language of the persons I would like to

thank.

Il primo ringraziamento va alla mia splendida, sgangherata famiglia greno-

blese, passata e presente: Lia, Ramo, Chiara, Gan, Vera, Paolo, GBB e Ste-

fania, Giova1, Simeon, Eric, Caro, Giova2, Ricky, Carlo e Clio+Panga, Cus,

Fil... Senza di voi sarei disperata (e probabilmente sotto un ponte (ˆ.-)). Siete

davvero la mia famiglia, e questa tesi esiste grazie a voi.

Je remercie encore une fois Francoise, pas dans son role de Directrice de these

mais plutot comme une chere amie qui a partage avec mois le travail de ces trois

annees. Merci pour tout, je n’ai pas les mots pour t’exprimer ma gratitude.

Merci aussi a Robert, l’extraordinaire mari de Francoise, pour sa gentillesse, son

esprit agreable, les conversations en italien et l’habilite infaillible de trouver les

meilleurs restos a San Francisco et a Helsinki. Merci a Sylvain qui a ete tou-

jours pret a m’aider quand j’en ai eu besoin, et qui ma toujours donne des bons

conseils. Merci a Fred, qui m’a aide beaucoup pendent mes premiers jours au

166

Acknowledgments

Leti. Merci encore a l’equipe PCM, formee des personnes tellement agreables

que travailler avec eux c’est un plaisir. Un merci et un biz a Pilou et a Geor-

gette! Je remerci beaucoup Robert et Ariel pour la gentillesse et les moments

de bonheur, et le tres sympa Jean-Yves sans lequel cette these n’aurait que trois

chapitres.

Merci encore a mes amies Audrey, Kavita, Sylvia, aux thesards du labo et du

projet, et merci aussi aux thesards DCOS ”collegues de Lia” pour les repas bien

passes. Merci a Radekko, Cornelia, Pablo et a tout les autres ”etrangers de

Grenoble” qui forment une communaute tres chaleureuse.

Ringrazio gli amici lasciati a Milano, che non mi hanno dimenticata in questi

tre anni: la Piera, che resta la mia insostituibile migliore amica, Deiv e i suoi

cappelli che amo immensamente, Albe, Eleele e Qc (e il terzo incomodo), Vera,

Noja, Rezzo, il Niggah, Vanish, Dade, Tumji e Sunday, la Brini, Matt e il sem-

pregiovane Frank. Un pensiero pieno di gratitudine va a Marco, che mi ha

accompagnata alla soglia di questo viaggio, e alla sua famiglia. Ringrazio con

tantissimo affetto le meravigliose amiche del giardino, Gloria, Rubina, Elisa,

Stefania e Tosca, che tra alti e bassi e nonostante la lontananza restano un

caposaldo fondamentale della mia vita. Grazie di cuore a tutta la famiglia Ar-

brun/Maurino, ed in particolare a Matilde, Ezio e Floriana, per avermi accolta

con tanto affetto.

Rigrazio con tutto il cuore il mio Stefano, compagno della difficile stesura di

questa tesi e compagno della mia vita, senza il quale nulla avrebbe senso. Gra-

zie per essere stato con me in questa avventura, e per tutte quelle che ancora ci

aspettano.

Il ringraziamento piu sentito e profondo va alla mia piccola ma solida famiglia:

Ernesto, Antonella, Nike, Serena e Gio, Angelo e Silvia, i miei nonni. In par-

ticolare, tutta la mia gratitudine va a mamma e papa. Il vostro amore, la

comprensione e il supporto che mi avete sempre dato sono la base su cui poggia

la mia vita.

167

Abstract

AbstractPhase Change Memories (PCM) are one of the best candidates for the next generation of non volatile

memories. A great research effort is still needed in order to optimize the properties of phase change (PC)

materials which are used in PCM devices. In particular, doping has been demonstrated to improve retention

in devices. Moreover, a study of the effect of scaling and interface material on PC materials properties is still

an open research field. In this context, the first part of the thesis is dedicated to investigate the local structure

of C or N doped amorphous GeTe. The impact of doping is observed experimentally with the appearance of a

new peak in the pair distribution function of doped GeTe, indicating the formation of a bond at a new distance

that is absent in the undoped amorphous material. The presence of new environments involving carbon and

nitrogen is confirmed through ab initio simulations. The subject of the second part of this thesis is the impact

of confinement on Ge2Sb2Te5 (GST) crystallization mechanism. Nano-sized clusters of GST have been made

by sputtering, deposited and then studied through X-ray diffraction using synchrotron radiation. The crys-

talline clusters experience a tensile strain that can be ascribed to the effect of the embedding Al2O3 matrix.

Their crystallization temperature has been found to be only 25C higher than the one observed for a thin

film of GST of 10 nm deposited under the same conditions. This result is positive for the future development

PCM because it indicates that the scaling effect on the crystallization temperature in phase change material

can be small. The third and last part of the thesis is dedicated to the investigation of the interface material

effect on the crystallization temperature of GeTe and GST thin films through reflectivity and X-ray diffraction

measurements. In both GeTe and GST film 100 nm thick interfaced with Ta the crystallization temperature

is higher than in the case of TiN or SiO2 interface. Such an interface effect on relatively thick films was never

reported before. The results suggest that the SiO2 /GeTe interface is energetically favorable for the nucleation

and growth of grains with a preferred orientation and that nucleation and growth mechanisms are different for

different interface materials.

ResumeLes memoires a changement de phase sont l’un des candidats les plus prometteurs pour la prochaine

generation de memoires non-volatiles. Un intense effort de recherche est requis pour optimiser les materiaux

a changement de phase (PC) utilises dans ces memoires. En particulier, il a ete demontre que le dopage

ameliore les proprietes de retention des dispositifs. Par ailleurs, l’etude des effets de reduction de taille et des

effets des materiaux d’interface sur les proprietes des materiaux a changement de phase est encore un sujet de

recherche ouvert. Dans ce contexte, la premiere partie de la these est dediee a l’investigation de la structure

locale de GeTe amorphe dope avec C ou N. L’effet du dopage sur la structure a ete observe experimentalement

via l’apparition d’un nouveau pic dans la fonction de distribution de paires de GeTe dope, ce qui montre

la formation d’une nouvelle liaison interatomique absente dans le materiau non dope. La presence de nou-

velles configurations incluant le carbone et l’azote a ete confirmee par des simulations ab initio. L’objet de la

deuxieme partie de la these est l’influence de la reduction de taille sur la cristallisation de Ge2Sb2Te5 (GST).

Des agregats nanometriques de GST ont ete fabriques par pulverisation puis deposes et etudies par diffrac-

tion des rayons X en utilisant le rayonnement synchrotron. Dans l’etat cristallise une tres forte deformation

positive des agregats est observee et attribuee a la matrice d’Al2O3 qui entoure les agregats. La temperature

de cristallisation des agregats est de 25C plus elevee que celle d’un film de GST de 10 nm depose dans les

memes conditions. Ce resultat est encourageant pour les futurs developpements des memoires a changement

de phase car il montre que l’effet de reduction de taille sur la temperature de cristallisation peut-etre faible.

La troisieme et derniere partie de la these est dediee a l’investigation des effets des materiaux d’interface sur la

temperature de cristallisation de films minces de GeTe et GST par des mesures de reflectivite et de diffraction

des rayons X. Pour les deux materiaux, la temperature de cristallisation de films de 100 nm est plus grande

pour une interface avec du Ta que pour une interface avec du TiN ou du SiO2 . Une difference aussi marquee

n’etait jamais montre auparavant. Les resultats suggerent que l’interface SiO2 /GeTe est energetiquement

favorable pour la nucleation et la croissance de grains avec une orientation preferentielle et que les mecanismes

de nucleation et croissance sont differents pour differents materiaux d’interface.

169

effets de dopage, de réduction de taille et d'interface

Documents